2025-06-26 16:12:03,164 - mmdet - INFO - Saving checkpoint at 12 epochs [ ] 0/81, elapsed: 0s, ETA:/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/torch/_tensor.py:575: UserWarning: floor_divide is deprecated, and will be removed in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor'). (Triggered internally at ../aten/src/ATen/native/BinaryOps.cpp:467.) return torch.floor_divide(self, other) [ ] 1/81, 0.3 task/s, elapsed: 3s, ETA: 273s/media/wangbaihui/1ecf654b-afad-4dab-af7b-e34b00dda87a/mmdetection3d/VAD/projects/mmdet3d_plugin/core/bbox/coders/fut_nms_free_coder.py:78: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor). self.post_center_range = torch.tensor( /media/wangbaihui/1ecf654b-afad-4dab-af7b-e34b00dda87a/mmdetection3d/VAD/projects/mmdet3d_plugin/core/bbox/coders/map_nms_free_coder.py:82: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor). self.post_center_range = torch.tensor( [>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>] 81/81, 4.5 task/s, elapsed: 18s, ETA: 0s Traceback (most recent call last): File "tools/train.py", line 266, in <module> main() File "tools/train.py", line 255, in main custom_train_model( File "/media/wangbaihui/1ecf654b-afad-4dab-af7b-e34b00dda87a/mmdetection3d/VAD/projects/mmdet3d_plugin/VAD/apis/train.py", line 21, in custom_train_model custom_train_detector( File "/media/wangbaihui/1ecf654b-afad-4dab-af7b-e34b00dda87a/mmdetection3d/VAD/projects/mmdet3d_plugin/VAD/apis/mmdet_train.py", line 194, in custom_train_detector runner.run(data_loaders, cfg.workflow) File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/mmcv/runner/epoch_based_runner.py", line 127, in run epoch_runner(data_loaders[i], **kwargs) File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/mmcv/runner/epoch_based_runner.py", line 54, in train self.call_hook('after_train_epoch') File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/mmcv/runner/base_runner.py", line 307, in call_hook getattr(hook, fn_name)(self) File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/mmcv/runner/hooks/evaluation.py", line 267, in after_train_epoch self._do_evaluate(runner) File "/media/wangbaihui/1ecf654b-afad-4dab-af7b-e34b00dda87a/mmdetection3d/VAD/projects/mmdet3d_plugin/core/evaluation/eval_hooks.py", line 88, in _do_evaluate key_score = self.evaluate(runner, results) File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/mmcv/runner/hooks/evaluation.py", line 361, in evaluate eval_res = self.dataloader.dataset.evaluate( File "/media/wangbaihui/1ecf654b-afad-4dab-af7b-e34b00dda87a/mmdetection3d/VAD/projects/mmdet3d_plugin/datasets/nuscenes_vad_dataset.py", line 1781, in evaluate all_metric_dict[key] += results[i]['metric_results'][key] KeyError: 0 ERROR:torch.distributed.elastic.multiprocessing.api:failed (exitcode: 1) local_rank: 0 (pid: 2522198) of binary: /home/wangbaihui/anaconda3/envs/vad/bin/python /home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py:367: UserWarning: ********************************************************************** CHILD PROCESS FAILED WITH NO ERROR_FILE ********************************************************************** CHILD PROCESS FAILED WITH NO ERROR_FILE Child process 2522198 (local_rank 0) FAILED (exitcode 1) Error msg: Process failed with exitcode 1 Without writing an error file to <N/A>. While this DOES NOT affect the correctness of your application, no trace information about the error will be available for inspection. Consider decorating your top level entrypoint function with torch.distributed.elastic.multiprocessing.errors.record. Example: from torch.distributed.elastic.multiprocessing.errors import record @record def trainer_main(args): # do train ********************************************************************** warnings.warn(_no_error_file_warning_msg(rank, failure)) Traceback (most recent call last): File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/runpy.py", line 194, in _run_module_as_main return _run_code(code, main_globals, None, File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/runpy.py", line 87, in _run_code exec(code, run_globals) File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/torch/distributed/run.py", line 702, in <module> main() File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 361, in wrapper return f(*args, **kwargs) File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/torch/distributed/run.py", line 698, in main run(args) File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/torch/distributed/run.py", line 689, in run elastic_launch( File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/torch/distributed/launcher/api.py", line 116, in __call__ return launch_agent(self._config, self._entrypoint, list(args)) File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/torch/distributed/launcher/api.py", line 244, in launch_agent raise ChildFailedError( torch.distributed.elastic.multiprocessing.errors.ChildFailedError: *************************************** tools/train.py FAILED ======================================= Root Cause: [0]: time: 2025-06-26_16:12:28 rank: 0 (local_rank: 0) exitcode: 1 (pid: 2522198) error_file: <N/A> msg: "Process failed with exitcode 1" ======================================= Other Failures: <NO_OTHER_FAILURES> ***********2025-06-26 16:12:03,164 - mmdet - INFO - Saving checkpoint at 12 epochs [ ] 0/81, elapsed: 0s, ETA:/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/torch/_tensor.py:575: UserWarning: floor_divide is deprecated, and will be removed in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor'). (Triggered internally at ../aten/src/ATen/native/BinaryOps.cpp:467.) return torch.floor_divide(self, other) [ ] 1/81, 0.3 task/s, elapsed: 3s, ETA: 273s/media/wangbaihui/1ecf654b-afad-4dab-af7b-e34b00dda87a/mmdetection3d/VAD/projects/mmdet3d_plugin/core/bbox/coders/fut_nms_free_coder.py:78: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor). self.post_center_range = torch.tensor( /media/wangbaihui/1ecf654b-afad-4dab-af7b-e34b00dda87a/mmdetection3d/VAD/projects/mmdet3d_plugin/core/bbox/coders/map_nms_free_coder.py:82: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor). self.post_center_range = torch.tensor( [>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>] 81/81, 4.5 task/s, elapsed: 18s, ETA: 0s Traceback (most recent call last): File "tools/train.py", line 266, in <module> main() File "tools/train.py", line 255, in main custom_train_model( File "/media/wangbaihui/1ecf654b-afad-4dab-af7b-e34b00dda87a/mmdetection3d/VAD/projects/mmdet3d_plugin/VAD/apis/train.py", line 21, in custom_train_model custom_train_detector( File "/media/wangbaihui/1ecf654b-afad-4dab-af7b-e34b00dda87a/mmdetection3d/VAD/projects/mmdet3d_plugin/VAD/apis/mmdet_train.py", line 194, in custom_train_detector runner.run(data_loaders, cfg.workflow) File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/mmcv/runner/epoch_based_runner.py", line 127, in run epoch_runner(data_loaders[i], **kwargs) File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/mmcv/runner/epoch_based_runner.py", line 54, in train self.call_hook('after_train_epoch') File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/mmcv/runner/base_runner.py", line 307, in call_hook getattr(hook, fn_name)(self) File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/mmcv/runner/hooks/evaluation.py", line 267, in after_train_epoch self._do_evaluate(runner) File "/media/wangbaihui/1ecf654b-afad-4dab-af7b-e34b00dda87a/mmdetection3d/VAD/projects/mmdet3d_plugin/core/evaluation/eval_hooks.py", line 88, in _do_evaluate key_score = self.evaluate(runner, results) File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/mmcv/runner/hooks/evaluation.py", line 361, in evaluate eval_res = self.dataloader.dataset.evaluate( File "/media/wangbaihui/1ecf654b-afad-4dab-af7b-e34b00dda87a/mmdetection3d/VAD/projects/mmdet3d_plugin/datasets/nuscenes_vad_dataset.py", line 1781, in evaluate all_metric_dict[key] += results[i]['metric_results'][key] KeyError: 0 ERROR:torch.distributed.elastic.multiprocessing.api:failed (exitcode: 1) local_rank: 0 (pid: 2522198) of binary: /home/wangbaihui/anaconda3/envs/vad/bin/python /home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py:367: UserWarning: ********************************************************************** CHILD PROCESS FAILED WITH NO ERROR_FILE ********************************************************************** CHILD PROCESS FAILED WITH NO ERROR_FILE Child process 2522198 (local_rank 0) FAILED (exitcode 1) Error msg: Process failed with exitcode 1 Without writing an error file to <N/A>. While this DOES NOT affect the correctness of your application, no trace information about the error will be available for inspection. Consider decorating your top level entrypoint function with torch.distributed.elastic.multiprocessing.errors.record. Example: from torch.distributed.elastic.multiprocessing.errors import record @record def trainer_main(args): # do train ********************************************************************** warnings.warn(_no_error_file_warning_msg(rank, failure)) Traceback (most recent call last): File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/runpy.py", line 194, in _run_module_as_main return _run_code(code, main_globals, None, File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/runpy.py", line 87, in _run_code exec(code, run_globals) File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/torch/distributed/run.py", line 702, in <module> main() File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 361, in wrapper return f(*args, **kwargs) File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/torch/distributed/run.py", line 698, in main run(args) File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/torch/distributed/run.py", line 689, in run elastic_launch( File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/torch/distributed/launcher/api.py", line 116, in __call__ return launch_agent(self._config, self._entrypoint, list(args)) File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/torch/distributed/launcher/api.py", line 244, in launch_agent raise ChildFailedError( torch.distributed.elastic.multiprocessing.errors.ChildFailedError: *************************************** tools/train.py FAILED ======================================= Root Cause: [0]: time: 2025-06-26_16:12:28 rank: 0 (local_rank: 0) exitcode: 1 (pid: 2522198) error_file: <N/A> msg: "Process failed with exitcode 1" ======================================= Other Failures: <NO_OTHER_FAILURES> ***********

时间: 2025-06-30 14:19:34 浏览: 32
在 PyTorch 分布式训练过程中出现 `KeyError: 0` 和 `ChildFailedError` 的问题，通常与分布式环境配置不正确或数据加载过程中的异常有关。以下是对这些问题的排查和修复建议： ### KeyError: 'RANK' 及相关环境变量缺失 在某些情况下，启动脚本未正确设置分布式训练所需的环境变量（如 `RANK`, `WORLD_SIZE`, `LOCAL_RANK`），导致程序运行时报错。PyTorch 分布式训练依赖这些环境变量来管理进程组和通信机制。如果使用 `torch.distributed.launch` 或 `torchrun` 启动训练脚本，则需要确保启动命令中包含必要的参数，并且脚本内部能够正确读取这些变量[^3]。 例如，正确的启动命令如下所示： ```bash python -m torch.distributed.launch --nproc_per_node=2 train.py ``` 其中 `--nproc_per_node` 指定了单机上的 GPU 数量，该参数会自动为每个进程分配 `LOCAL_RANK` 环境变量。此外，在脚本中应通过 `argparse` 解析 `--local_rank` 参数并传递给 `torch.distributed.init_process_group`，以完成分布式初始化流程[^4]。 ### KeyError: 0 与 NCCL 初始化失败 当使用多机多卡训练时，可能会遇到 `KeyError: 0` 错误，这通常是由 `c10d` key-value store 在获取 `ncclUniqueId` 时失败引起的。此错误可能表明节点之间的网络通信存在问题，或者 NCCL 初始化未能成功执行。可以通过检查以下几点进行排查： - **网络连接**：确保所有节点之间可以通过指定端口通信，且防火墙规则未阻止相关流量。 - **IP 地址和端口一致性**：确认主节点 IP 地址和通信端口在所有进程中一致，并且在调用 `init_process_group` 时正确设置。 - **NCCL 版本兼容性**：检查使用的 NCCL 版本是否与当前 CUDA 和 PyTorch 版本兼容，必要时升级 NCCL 或调整 PyTorch 版本。 - **资源限制**：确保系统没有达到文件描述符、内存或其他资源的上限，这些也可能影响 NCCL 初始化[^2]。 ### ChildFailedError 与子进程崩溃 `ChildFailedError` 表示某个子进程在启动后立即崩溃。这类问题通常由以下原因引起： - **CUDA 初始化失败**：检查设备是否可用，GPU 驱动是否安装正确，以及 CUDA 版本是否与 PyTorch 兼容。 - **数据加载器问题**：如果在数据加载阶段发生错误，可能导致子进程提前终止。可以尝试将 `num_workers` 设置为 0 来排除 DataLoader 中的问题。 - **代码逻辑错误**：某些特定于进程的操作（如模型初始化、数据分片）可能在某个进程中触发异常。可以在子进程中加入日志记录或断言检查，定位具体出错位置。 ### 数据处理中的 KeyError: '000164' 在 Opencood 项目中，训练过程中出现偶数编号的 `KeyError: '000164'` 报错，可能是由于数据索引或标注文件解析错误所致。这种现象通常出现在自定义数据集加载器中，尤其是在构建样本路径或读取标注信息时未能正确处理部分条目。建议采取以下措施： - **检查数据索引一致性**：确保数据集索引列表完整，不存在缺失或重复的 ID。 - **验证标注文件格式**：确认标注文件结构与代码预期一致，特别是字段名称和类型。 - **添加异常捕获机制**：在数据加载函数中加入 `try-except` 块，跳过有问题的数据项并输出警告信息，以便继续训练[^1]。 ### 示例代码：修改数据加载器以忽略错误数据项 ```python def __getitem__(self, index): try: # 正常数据加载逻辑 return self._load_data(index) except KeyError as e: print(f"Skipping corrupted data at index {index}: {e}") return self.__getitem__((index + 1) % len(self)) ``` 上述方法可在不影响整体训练流程的前提下，临时绕过损坏或格式不正确的数据项。
阅读全文

相关推荐

zip

dlr-checkpoint-navigation-controller-ios:底特律实验室食谱 - 检查点导航控制器

DLR检查点导航控制器 安装 DLRCheckpointNavigationController 可通过。 要安装它，只需将以下行添加到您的 Podfile 中： pod "DLRCheckpointNavigationController" 作者 内森·， 执照 ... 有关详细信息，请参阅...
zip

MobileNetV3-SSD-Compact-Version:MobileNetV3 SSD的简洁版本

MobileNetV3-SSD-Compact-Version MobileNetV3 SSD的简洁...一 下载VOC数据集之后,将VOCtrainval_06-Nov-2007和VOCtest_06-Nov-2007合并在一起 数据集下载 可以看这里 二 先打开create_data_lists.py文件 改成自己数
zip

uPIT-for-speech-separation:语音级别的PIT实验进行语音分离

使用uPIT进行语音分离使用发声级PIT（置换不变训练）进行语音分离要求参见用法使用生成数据集准备cmvn，.scp并在.yaml文件中配置实验训练： ./run_pit.py --config $conf --num-epoches 100 > $checkpoint /train....

25-06-26 14:46:11,967 - mmdet - INFO - Saving checkpoint at 48 epochs [ ] 0/81, elapsed: 0s, ETA:/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/torch/_tensor.py:575: UserWarning: floor_divide is deprecated, and will be removed in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor'). (Triggered internally at ../aten/src/ATen/native/BinaryOps.cpp:467.) return torch.floor_divide(self, other) [ ] 1/81, 0.3 task/s, elapsed: 4s, ETA: 303s/media/wangbaihui/1ecf654b-afad-4dab-af7b-e34b00dda87a/mmdetection3d/VAD/projects/mmdet3d_plugin/core/bbox/coders/fut_nms_free_coder.py:78: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor). self.post_center_range = torch.tensor( /media/wangbaihui/1ecf654b-afad-4dab-af7b-e34b00dda87a/mmdetection3d/VAD/projects/mmdet3d_plugin/core/bbox/coders/map_nms_free_coder.py:82: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor). self.post_center_range = torch.tensor( [>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>] 81/81, 3.4 task/s, elapsed: 24s, ETA: 0s Traceback (most recent call last): File "tools/train.py", line 266, in <module> main() File "tools/train.py", line 255, in main custom_train_model( File "/media/wangbaihui/1ecf654b-afad-4dab-af7b-e34b00dda87a/mmdetection3d/VAD/projects/mmdet3d_plugin/VAD/apis/train.py", line 21, in custom_train_model custom_train_detector( File "/media/wangbaihui/1ecf654b-afad-4dab-af7b-e34b00dda87a/mmdetection3d/VAD/projects/mmdet3d_plugin/VAD/apis/mmdet_train.py", line 194, in custom_train_detector runner.run(data_loaders, cfg.workflow) File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/mmcv/runner/epoch_based_runner.py", line 127, in run epoch_runner(data_loaders[i], **kwargs) File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/mmcv/runner/epoch_based_runner.py", line 54, in train self.call_hook('after_train_epoch') File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/mmcv/runner/base_runner.py", line 307, in call_hook getattr(hook, fn_name)(self) File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/mmcv/runner/hooks/evaluation.py", line 267, in after_train_epoch self._do_evaluate(runner) File "/media/wangbaihui/1ecf654b-afad-4dab-af7b-e34b00dda87a/mmdetection3d/VAD/projects/mmdet3d_plugin/core/evaluation/eval_hooks.py", line 88, in _do_evaluate key_score = self.evaluate(runner, results) File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/mmcv/runner/hooks/evaluation.py", line 361, in evaluate eval_res = self.dataloader.dataset.evaluate( File "/media/wangbaihui/1ecf654b-afad-4dab-af7b-e34b00dda87a/mmdetection3d/VAD/projects/mmdet3d_plugin/datasets/nuscenes_vad_dataset.py", line 1781, in evaluate all_metric_dict[key] += results[i]['metric_results'][key] KeyError: 0 ERROR:torch.distributed.elastic.multiprocessing.api:failed (exitcode: 1) local_rank: 0 (pid: 2469743) of binary: /home/wangbaihui/anaconda3/envs/vad/bin/python /home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py:367: UserWarning: ********************************************************************** CHILD PROCESS FAILED WITH NO ERROR_FILE ********************************************************************** CHILD PROCESS FAILED WITH NO ERROR_FILE Child process 2469743 (local_rank 0) FAILED (exitcode 1) Error msg: Process failed with exitcode 1 Without writing an error file to <N/A>. While this DOES NOT affect the correctness of your application, no trace information about the error will be available for inspection. Consider decorating your top level entrypoint function with torch.distributed.elastic.multiprocessing.errors.record. Example: from torch.distributed.elastic.multiprocessing.errors import record @record def trainer_main(args): # do train ********************************************************************** warnings.warn(_no_error_file_warning_msg(rank, failure)) Traceback (most recent call last): File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/runpy.py", line 194, in _run_module_as_main return _run_code(code, main_globals, None, File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/runpy.py", line 87, in _run_code exec(code, run_globals) File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/torch/distributed/run.py", line 702, in <module> main() File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 361, in wrapper return f(*args, **kwargs) File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/torch/distributed/run.py", line 698, in main run(args) File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/torch/distributed/run.py", line 689, in run elastic_launch( File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/torch/distributed/launcher/api.py", line 116, in __call__ return launch_agent(self._config, self._entrypoint, list(args)) File "/home/wangbaihui/anaconda3/envs/vad/lib/python3.8/site-packages/torch/distributed/launcher/api.py", line 244, in launch_agent raise ChildFailedError( torch.distributed.elastic.multiprocessing.errors.ChildFailedError: *************************************** tools/train.py FAILED ======================================= Root Cause: [0]: time: 2025-06-26_14:46:41 rank: 0 (local_rank: 0) exitcode: 1 (pid: 2469743) error_file: <N/A> msg: "Process failed with exitcode 1" ======================================= Other Failures: <NO_OTHER_FAILURES>

我们正在使用mmdetection3d进行训练，在评估阶段遇到了KeyError:0的错误。根据引用[3]中的经验，这种错误通常与评估配置有关。在mmdetection中，评估器（evaluator）的配置决定了计算哪些指标（如bbox、segm等）。...

from data import * from utils.augmentations import SSDAugmentation, BaseTransform from utils.functions import MovingAverage, SavePath from utils.logger import Log from utils import timer from layers.modules import MultiBoxLoss from yolact import Yolact import os import sys import time import math, random from pathlib import Path import torch from torch.autograd import Variable import torch.nn as nn import torch.optim as optim import torch.backends.cudnn as cudnn import torch.nn.init as init import torch.utils.data as data import numpy as np import argparse import datetime # Oof import eval as eval_script def str2bool(v): return v.lower() in ("yes", "true", "t", "1") parser = argparse.ArgumentParser( description='Yolact Training Script') parser.add_argument('--batch_size', default=2, type=int, help='Batch size for training') parser.add_argument('--resume', default=None, type=str, help='Checkpoint state_dict file to resume training from. If this is "interrupt"'\ ', the model will resume training from the interrupt file.') parser.add_argument('--start_iter', default=-1, type=int, help='Resume training at this iter. If this is -1, the iteration will be'\ 'determined from the file name.') parser.add_argument('--num_workers', default=0, type=int, help='Number of workers used in dataloading') parser.add_argument('--cuda', default=True, type=str2bool, help='Use CUDA to train model') parser.add_argument('--lr', '--learning_rate', default=None, type=float, help='Initial learning rate. Leave as None to read this from the config.') parser.add_argument('--momentum', default=None, type=float, help='Momentum for SGD. Leave as None to read this from the config.') parser.add_argument('--decay', '--weight_decay', default=None, type=float, help='Weight decay for SGD. Leave as None to read this from the config.') parser.add_argument('--gamma', default=None, type=float, help='For each lr step, what to multiply the lr by. Leave as None to read this from the config.') parser.add_argument('--save_folder', default='weights/', help='Directory for saving checkpoint models.') parser.add_argument('--log_folder', default='logs/', help='Directory for saving logs.') parser.add_argument('--config', default=None, help='The config object to use.') parser.add_argument('--save_interval', default=10000, type=int, help='The number of iterations between saving the model.') parser.add_argument('--validation_size', default=5000, type=int, help='The number of images to use for validation.') parser.add_argument('--validation_epoch', default=2, type=int, help='Output validation information every n iterations. If -1, do no validation.') parser.add_argument('--keep_latest', dest='keep_latest', action='store_true', help='Only keep the latest checkpoint instead of each one.') parser.add_argument('--keep_latest_interval', default=100000, type=int, help='When --keep_latest is on, don\'t delete the latest file at these intervals. This should be a multiple of save_interval or 0.') parser.add_argument('--dataset', default=None, type=str, help='If specified, override the dataset specified in the config with this one (example: coco2017_dataset).') parser.add_argument('--no_log', dest='log', action='store_false', help='Don\'t log per iteration information into log_folder.') parser.add_argument('--log_gpu', dest='log_gpu', action='store_true', help='Include GPU information in the logs. Nvidia-smi tends to be slow, so set this with caution.') parser.add_argument('--no_interrupt', dest='interrupt', action='store_false', help='Don\'t save an interrupt when KeyboardInterrupt is caught.') parser.add_argument('--batch_alloc', default=None, type=str, help='If using multiple GPUS, you can set this to be a comma separated list detailing which GPUs should get what local batch size (It should add up to your total batch size).') parser.add_argument('--no_autoscale', dest='autoscale', action='store_false', help='YOLACT will automatically scale the lr and the number of iterations depending on the batch size. Set this if you want to disable that.') parser.set_defaults(keep_latest=False, log=True, log_gpu=False, interrupt=True, autoscale=True) args = parser.parse_args() if args.config is not None: set_cfg(args.config) if args.dataset is not None: set_dataset(args.dataset) if args.autoscale and args.batch_size != 8: factor = args.batch_size / 8 if __name__ == '__main__': print('Scaling parameters by %.2f to account for a batch size of %d.' % (factor, args.batch_size)) cfg.lr *= factor cfg.max_iter //= factor cfg.lr_steps = [x // factor for x in cfg.lr_steps] # Update training parameters from the config if necessary def replace(name): if getattr(args, name) == None: setattr(args, name, getattr(cfg, name)) replace('lr') replace('decay') replace('gamma') replace('momentum') # This is managed by set_lr cur_lr = args.lr if torch.cuda.device_count() == 0: print('No GPUs detected. Exiting...') exit(-1) if args.batch_size // torch.cuda.device_count() < 6: if __name__ == '__main__': print('Per-GPU batch size is less than the recommended limit for batch norm. Disabling batch norm.') cfg.freeze_bn = True loss_types = ['B', 'C', 'M', 'P', 'D', 'E', 'S', 'I'] if torch.cuda.is_available(): if args.cuda: torch.set_default_tensor_type('torch.cuda.FloatTensor') if not args.cuda: print("WARNING: It looks like you have a CUDA device, but aren't " + "using CUDA.\nRun with --cuda for optimal training speed.") torch.set_default_tensor_type('torch.FloatTensor') else: torch.set_default_tensor_type('torch.FloatTensor') class NetLoss(nn.Module): """ A wrapper for running the network and computing the loss This is so we can more efficiently use DataParallel. """ def __init__(self, net:Yolact, criterion:MultiBoxLoss): super().__init__() self.net = net self.criterion = criterion def forward(self, images, targets, masks, num_crowds): preds = self.net(images) losses = self.criterion(self.net, preds, targets, masks, num_crowds) return losses class CustomDataParallel(nn.DataParallel): """ This is a custom version of DataParallel that works better with our training data. It should also be faster than the general case. """ def scatter(self, inputs, kwargs, device_ids): # More like scatter and data prep at the same time. The point is we prep the data in such a way # that no scatter is necessary, and there's no need to shuffle stuff around different GPUs. devices = ['cuda:' + str(x) for x in device_ids] splits = prepare_data(inputs[0], devices, allocation=args.batch_alloc) return [[split[device_idx] for split in splits] for device_idx in range(len(devices))], \ [kwargs] * len(devices) def gather(self, outputs, output_device): out = {} for k in outputs[0]: out[k] = torch.stack([output[k].to(output_device) for output in outputs]) return out def train(): if not os.path.exists(args.save_folder): os.mkdir(args.save_folder) dataset = COCODetection(image_path=cfg.dataset.train_images, info_file=cfg.dataset.train_info, transform=SSDAugmentation(MEANS)) if args.validation_epoch > 0: setup_eval() val_dataset = COCODetection(image_path=cfg.dataset.valid_images, info_file=cfg.dataset.valid_info, transform=BaseTransform(MEANS)) # Parallel wraps the underlying module, but when saving and loading we don't want that yolact_net = Yolact() net = yolact_net net.train() if args.log: log = Log(cfg.name, args.log_folder, dict(args._get_kwargs()), overwrite=(args.resume is None), log_gpu_stats=args.log_gpu) # I don't use the timer during training (I use a different timing method). # Apparently there's a race condition with multiple GPUs, so disable it just to be safe. timer.disable_all() # Both of these can set args.resume to None, so do them before the check if args.resume == 'interrupt': args.resume = SavePath.get_interrupt(args.save_folder) elif args.resume == 'latest': args.resume = SavePath.get_latest(args.save_folder, cfg.name) if args.resume is not None: print('Resuming training, loading {}...'.format(args.resume)) yolact_net.load_weights(args.resume) if args.start_iter == -1: args.start_iter = SavePath.from_str(args.resume).iteration else: print('Initializing weights...') yolact_net.init_weights(backbone_path=args.save_folder + cfg.backbone.path) optimizer = optim.SGD(net.parameters(), lr=args.lr, momentum=args.momentum, weight_decay=args.decay) criterion = MultiBoxLoss(num_classes=cfg.num_classes, pos_threshold=cfg.positive_iou_threshold, neg_threshold=cfg.negative_iou_threshold, negpos_ratio=cfg.ohem_negpos_ratio) if args.batch_alloc is not None: args.batch_alloc = [int(x) for x in args.batch_alloc.split(',')] if sum(args.batch_alloc) != args.batch_size: print('Error: Batch allocation (%s) does not sum to batch size (%s).' % (args.batch_alloc, args.batch_size)) exit(-1) net = CustomDataParallel(NetLoss(net, criterion)) if args.cuda: net = net.cuda() # Initialize everything if not cfg.freeze_bn: yolact_net.freeze_bn() # Freeze bn so we don't kill our means yolact_net(torch.zeros(1, 3, cfg.max_size, cfg.max_size).cuda()) if not cfg.freeze_bn: yolact_net.freeze_bn(True) # loss counters loc_loss = 0 conf_loss = 0 iteration = max(args.start_iter, 0) last_time = time.time() epoch_size = len(dataset)+1 // args.batch_size num_epochs = math.ceil(cfg.max_iter / epoch_size) # Which learning rate adjustment step are we on? lr' = lr * gamma ^ step_index step_index = 0 data_loader = data.DataLoader(dataset, args.batch_size, num_workers=args.num_workers, shuffle=True, collate_fn=detection_collate, pin_memory=True) save_path = lambda epoch, iteration: SavePath(cfg.name, epoch, iteration).get_path(root=args.save_folder) time_avg = MovingAverage() global loss_types # Forms the print order loss_avgs = { k: MovingAverage(100) for k in loss_types } print('Begin training!') print() # try-except so you can use ctrl+c to save early and stop training try: for epoch in range(num_epochs): # Resume from start_iter if (epoch+1)*epoch_size < iteration: continue for datum in data_loader: # Stop if we've reached an epoch if we're resuming from start_iter if iteration == (epoch+1)*epoch_size: break # Stop at the configured number of iterations even if mid-epoch if iteration == cfg.max_iter: break # Change a config setting if we've reached the specified iteration changed = False for change in cfg.delayed_settings: if iteration >= change[0]: changed = True cfg.replace(change[1]) # Reset the loss averages because things might have changed for avg in loss_avgs: avg.reset() # If a config setting was changed, remove it from the list so we don't keep checking if changed: cfg.delayed_settings = [x for x in cfg.delayed_settings if x[0] > iteration] # Warm up by linearly interpolating the learning rate from some smaller value if cfg.lr_warmup_until > 0 and iteration <= cfg.lr_warmup_until: set_lr(optimizer, (args.lr - cfg.lr_warmup_init) * (iteration / cfg.lr_warmup_until) + cfg.lr_warmup_init) # Adjust the learning rate at the given iterations, but also if we resume from past that iteration while step_index < len(cfg.lr_steps) and iteration >= cfg.lr_steps[step_index]: step_index += 1 set_lr(optimizer, args.lr * (args.gamma ** step_index)) # Zero the grad to get ready to compute gradients optimizer.zero_grad() # Forward Pass + Compute loss at the same time (see CustomDataParallel and NetLoss) losses = net(datum) losses = { k: (v).mean() for k,v in losses.items() } # Mean here because Dataparallel loss = sum([losses[k] for k in losses]) # no_inf_mean removes some components from the loss, so make sure to backward through all of it # all_loss = sum([v.mean() for v in losses.values()]) # Backprop loss.backward() # Do this to free up vram even if loss is not finite if torch.isfinite(loss).item(): optimizer.step() # Add the loss to the moving average for bookkeeping for k in losses: loss_avgs[k].add(losses[k].item()) cur_time = time.time() elapsed = cur_time - last_time last_time = cur_time # Exclude graph setup from the timing information if iteration != args.start_iter: time_avg.add(elapsed) if iteration % 10 == 0: eta_str = str(datetime.timedelta(seconds=(cfg.max_iter-iteration) * time_avg.get_avg())).split('.')[0] total = sum([loss_avgs[k].get_avg() for k in losses]) loss_labels = sum([[k, loss_avgs[k].get_avg()] for k in loss_types if k in losses], []) print(('[%3d] %7d ||' + (' %s: %.3f |' * len(losses)) + ' T: %.3f || ETA: %s || timer: %.3f') % tuple([epoch, iteration] + loss_labels + [total, eta_str, elapsed]), flush=True) if args.log: precision = 5 loss_info = {k: round(losses[k].item(), precision) for k in losses} loss_info['T'] = round(loss.item(), precision) if args.log_gpu: log.log_gpu_stats = (iteration % 10 == 0) # nvidia-smi is sloooow log.log('train', loss=loss_info, epoch=epoch, iter=iteration, lr=round(cur_lr, 10), elapsed=elapsed) log.log_gpu_stats = args.log_gpu iteration += 1 if iteration % args.save_interval == 0 and iteration != args.start_iter: if args.keep_latest: latest = SavePath.get_latest(args.save_folder, cfg.name) print('Saving state, iter:', iteration) yolact_net.save_weights(save_path(epoch, iteration)) if args.keep_latest and latest is not None: if args.keep_latest_interval <= 0 or iteration % args.keep_latest_interval != args.save_interval: print('Deleting old save...') os.remove(latest) # This is done per epoch if args.validation_epoch > 0: if epoch % args.validation_epoch == 0 and epoch > 0: compute_validation_map(epoch, iteration, yolact_net, val_dataset, log if args.log else None) # Compute validation mAP after training is finished compute_validation_map(epoch, iteration, yolact_net, val_dataset, log if args.log else None) except KeyboardInterrupt: if args.interrupt: print('Stopping early. Saving network...') # Delete previous copy of the interrupted network so we don't spam the weights folder SavePath.remove_interrupt(args.save_folder) yolact_net.save_weights(save_path(epoch, repr(iteration) + '_interrupt')) exit() yolact_net.save_weights(save_path(epoch, iteration)) def set_lr(optimizer, new_lr): for param_group in optimizer.param_groups: param_group['lr'] = new_lr global cur_lr cur_lr = new_lr def gradinator(x): x.requires_grad = False return x def prepare_data(datum, devices:list=None, allocation:list=None): with torch.no_grad(): if devices is None: devices = ['cuda:0'] if args.cuda else ['cpu'] if allocation is None: allocation = [args.batch_size // len(devices)] * (len(devices) - 1) allocation.append(args.batch_size - sum(allocation)) # The rest might need more/less images, (targets, masks, num_crowds) = datum cur_idx = 0 for device, alloc in zip(devices, allocation): for _ in range(alloc): images[cur_idx] = gradinator(images[cur_idx].to(device)) targets[cur_idx] = gradinator(targets[cur_idx].to(device)) masks[cur_idx] = gradinator(masks[cur_idx].to(device)) cur_idx += 1 if cfg.preserve_aspect_ratio: # Choose a random size from the batch _, h, w = images[random.randint(0, len(images)-1)].size() for idx, (image, target, mask, num_crowd) in enumerate(zip(images, targets, masks, num_crowds)): images[idx], targets[idx], masks[idx], num_crowds[idx] \ = enforce_size(image, target, mask, num_crowd, w, h) cur_idx = 0 split_images, split_targets, split_masks, split_numcrowds \ = [[None for alloc in allocation] for _ in range(4)] for device_idx, alloc in enumerate(allocation): split_images[device_idx] = torch.stack(images[cur_idx:cur_idx+alloc], dim=0) split_targets[device_idx] = targets[cur_idx:cur_idx+alloc] split_masks[device_idx] = masks[cur_idx:cur_idx+alloc] split_numcrowds[device_idx] = num_crowds[cur_idx:cur_idx+alloc] cur_idx += alloc return split_images, split_targets, split_masks, split_numcrowds def no_inf_mean(x:torch.Tensor): """ Computes the mean of a vector, throwing out all inf values. If there are no non-inf values, this will return inf (i.e., just the normal mean). """ no_inf = [a for a in x if torch.isfinite(a)] if len(no_inf) > 0: return sum(no_inf) / len(no_inf) else: return x.mean() def compute_validation_loss(net, data_loader, criterion): global loss_types with torch.no_grad(): losses = {} # Don't switch to eval mode because we want to get losses iterations = 0 for datum in data_loader: images, targets, masks, num_crowds = prepare_data(datum) out = net(images) wrapper = ScatterWrapper(targets, masks, num_crowds) _losses = criterion(out, wrapper, wrapper.make_mask()) for k, v in _losses.items(): v = v.mean().item() if k in losses: losses[k] += v else: losses[k] = v iterations += 1 if args.validation_size <= iterations * args.batch_size: break for k in losses: losses[k] /= iterations loss_labels = sum([[k, losses[k]] for k in loss_types if k in losses], []) print(('Validation ||' + (' %s: %.3f |' * len(losses)) + ')') % tuple(loss_labels), flush=True) def compute_validation_map(epoch, iteration, yolact_net, dataset, log:Log=None): with torch.no_grad(): yolact_net.eval() start = time.time() print() print("Computing validation mAP (this may take a while)...", flush=True) val_info = eval_script.evaluate(yolact_net, dataset, train_mode=True) end = time.time() if log is not None: log.log('val', val_info, elapsed=(end - start), epoch=epoch, iter=iteration) yolact_net.train() def setup_eval(): eval_script.parse_args(['--no_bar', '--max_images='+str(args.validation_size)]) if __name__ == '__main__': train() 上述代码怎么插进def count_parameters(model): return sum(p.numel() for p in model.parameters() if p.requires_grad) # 在模型初始化后调用 total_params = count_parameters(yolact_model) print(f"Params: {total_params/1e6:.3f}M") # 转换为百万单位 pip install thop # 先安装依赖 from thop import profile # 在模型初始化后添加 input = torch.randn(1, 3, cfg.img_size, cfg.img_size).to(device) flops, _ = profile(yolact_model, inputs=(input,)) print(f"GFLOPs: {flops/1e9:.2f}G") # 转换为十亿单位 import time # 在推理循环前添加 total_time = 0 num_frames = 100 # 测试帧数 # 预热 for _ in range(10): yolact_model(torch.randn(1,3,550,550).to(device)) # 正式测试 for i in range(num_frames): start = time.perf_counter() # 原推理代码 with torch.no_grad(): preds = yolact_model(input_image) # 后处理代码 # ... total_time += time.perf_counter() - start fps = num_frames / total_time print(f"FPS: {fps:.2f}") def calculate_iou(box1, box2): # box格式: [x1, y1, x2, y2] inter_x1 = max(box1[0], box2[0]) inter_y1 = max(box1[1], box2[1]) inter_x2 = min(box1[2], box2[2]) inter_y2 = min(box1[3], box2[3]) inter_area = max(0, inter_x2 - inter_x1) * max(0, inter_y2 - inter_y1) union_area = (box1[2]-box1[0])*(box1[3]-box1[1]) + \ (box2[2]-box2[0])*(box2[3]-box2[1]) - inter_area return inter_area / union_area # 在检测结果与GT匹配时调用 for pred_box, gt_box in zip(predictions, ground_truths): iou = calculate_iou(pred_box, gt_box) # 存储或输出iou def print_metrics(params, gflops, fps, iou=None): print(f"Model Metrics:") print(f"├── Parameters: {params:.3f}M") print(f"├── GFLOPs: {gflops:.2f}G") print(f"├── FPS: {fps:.2f}") if iou is not None: print(f"└── mIoU: {iou:.4f}") # 在评估流程结束时调用

print(f"└── mIoU: {val_info.get('mIoU', 0):.4f}") # 记录所有指标 if log is not None: metrics = { 'params': total_params, 'gflops': flops/1e9, 'fps': fps, 'mIoU': val_info.get('mIoU', 0) } ...

import torch from thop import profile # 1. 创建原始模型（未编译） model = YourModel().eval() # 确保在评估模式 # 2. 计算 FLOPs（thop 可安全添加 total_ops） input_tensor = torch.randn(1, 3, 224, 224) flops, params = profile(model, inputs=(input_tensor,)) # 3. 打印结果后编译模型 print(f"FLOPs: {flops/1e9:.2f} G, Params: {params/1e6:.2f} M") scripted_model = torch.jit.script(model) # 安全编译 这个代码放在以下代码哪个地方 from data import * from utils.augmentations import SSDAugmentation, BaseTransform from utils.functions import MovingAverage, SavePath from utils.logger import Log from utils import timer from layers.modules import MultiBoxLoss from yolact import Yolact from thop import profile import os import sys import time import math, random from pathlib import Path import torch from torch.autograd import Variable import torch.nn as nn import torch.optim as optim import torch.backends.cudnn as cudnn import torch.nn.init as init import torch.utils.data as data import numpy as np import argparse import datetime # Oof import eval as eval_script def str2bool(v): return v.lower() in ("yes", "true", "t", "1") parser = argparse.ArgumentParser( description='Yolact Training Script') parser.add_argument('--batch_size', default=2, type=int, help='Batch size for training') parser.add_argument('--resume', default=None, type=str, help='Checkpoint state_dict file to resume training from. If this is "interrupt"'\ ', the model will resume training from the interrupt file.') parser.add_argument('--start_iter', default=-1, type=int, help='Resume training at this iter. If this is -1, the iteration will be'\ 'determined from the file name.') parser.add_argument('--num_workers', default=0, type=int, help='Number of workers used in dataloading') parser.add_argument('--cuda', default=True, type=str2bool, help='Use CUDA to train model') parser.add_argument('--lr', '--learning_rate', default=None, type=float, help='Initial learning rate. Leave as None to read this from the config.') parser.add_argument('--momentum', default=None, type=float, help='Momentum for SGD. Leave as None to read this from the config.') parser.add_argument('--decay', '--weight_decay', default=None, type=float, help='Weight decay for SGD. Leave as None to read this from the config.') parser.add_argument('--gamma', default=None, type=float, help='For each lr step, what to multiply the lr by. Leave as None to read this from the config.') parser.add_argument('--save_folder', default='weights/', help='Directory for saving checkpoint models.') parser.add_argument('--log_folder', default='logs/', help='Directory for saving logs.') parser.add_argument('--config', default=None, help='The config object to use.') parser.add_argument('--save_interval', default=10000, type=int, help='The number of iterations between saving the model.') parser.add_argument('--validation_size', default=5000, type=int, help='The number of images to use for validation.') parser.add_argument('--validation_epoch', default=2, type=int, help='Output validation information every n iterations. If -1, do no validation.') parser.add_argument('--keep_latest', dest='keep_latest', action='store_true', help='Only keep the latest checkpoint instead of each one.') parser.add_argument('--keep_latest_interval', default=100000, type=int, help='When --keep_latest is on, don\'t delete the latest file at these intervals. This should be a multiple of save_interval or 0.') parser.add_argument('--dataset', default=None, type=str, help='If specified, override the dataset specified in the config with this one (example: coco2017_dataset).') parser.add_argument('--no_log', dest='log', action='store_false', help='Don\'t log per iteration information into log_folder.') parser.add_argument('--log_gpu', dest='log_gpu', action='store_true', help='Include GPU information in the logs. Nvidia-smi tends to be slow, so set this with caution.') parser.add_argument('--no_interrupt', dest='interrupt', action='store_false', help='Don\'t save an interrupt when KeyboardInterrupt is caught.') parser.add_argument('--batch_alloc', default=None, type=str, help='If using multiple GPUS, you can set this to be a comma separated list detailing which GPUs should get what local batch size (It should add up to your total batch size).') parser.add_argument('--no_autoscale', dest='autoscale', action='store_false', help='YOLACT will automatically scale the lr and the number of iterations depending on the batch size. Set this if you want to disable that.') parser.set_defaults(keep_latest=False, log=True, log_gpu=False, interrupt=True, autoscale=True) args = parser.parse_args() if args.config is not None: set_cfg(args.config) if args.dataset is not None: set_dataset(args.dataset) if args.autoscale and args.batch_size != 8: factor = args.batch_size / 8 if __name__ == '__main__': print('Scaling parameters by %.2f to account for a batch size of %d.' % (factor, args.batch_size)) cfg.lr *= factor cfg.max_iter //= factor cfg.lr_steps = [x // factor for x in cfg.lr_steps] # Update training parameters from the config if necessary def replace(name): if getattr(args, name) == None: setattr(args, name, getattr(cfg, name)) replace('lr') replace('decay') replace('gamma') replace('momentum') # This is managed by set_lr cur_lr = args.lr if torch.cuda.device_count() == 0: print('No GPUs detected. Exiting...') exit(-1) if args.batch_size // torch.cuda.device_count() < 6: if __name__ == '__main__': print('Per-GPU batch size is less than the recommended limit for batch norm. Disabling batch norm.') cfg.freeze_bn = True loss_types = ['B', 'C', 'M', 'P', 'D', 'E', 'S', 'I'] if torch.cuda.is_available(): if args.cuda: torch.set_default_tensor_type('torch.cuda.FloatTensor') if not args.cuda: print("WARNING: It looks like you have a CUDA device, but aren't " + "using CUDA.\nRun with --cuda for optimal training speed.") torch.set_default_tensor_type('torch.FloatTensor') else: torch.set_default_tensor_type('torch.FloatTensor') class NetLoss(nn.Module): """ A wrapper for running the network and computing the loss This is so we can more efficiently use DataParallel. """ def __init__(self, net:Yolact, criterion:MultiBoxLoss): super().__init__() self.net = net self.criterion = criterion def forward(self, images, targets, masks, num_crowds): preds = self.net(images) losses = self.criterion(self.net, preds, targets, masks, num_crowds) return losses class CustomDataParallel(nn.DataParallel): """ This is a custom version of DataParallel that works better with our training data. It should also be faster than the general case. """ def scatter(self, inputs, kwargs, device_ids): # More like scatter and data prep at the same time. The point is we prep the data in such a way # that no scatter is necessary, and there's no need to shuffle stuff around different GPUs. devices = ['cuda:' + str(x) for x in device_ids] splits = prepare_data(inputs[0], devices, allocation=args.batch_alloc) return [[split[device_idx] for split in splits] for device_idx in range(len(devices))], \ [kwargs] * len(devices) def gather(self, outputs, output_device): out = {} for k in outputs[0]: out[k] = torch.stack([output[k].to(output_device) for output in outputs]) return out def train(): if not os.path.exists(args.save_folder): os.mkdir(args.save_folder) dataset = COCODetection(image_path=cfg.dataset.train_images, info_file=cfg.dataset.train_info, transform=SSDAugmentation(MEANS)) if args.validation_epoch > 0: setup_eval() val_dataset = COCODetection(image_path=cfg.dataset.valid_images, info_file=cfg.dataset.valid_info, transform=BaseTransform(MEANS)) # Parallel wraps the underlying module, but when saving and loading we don't want that yolact_net = Yolact() net = yolact_net net.train() # 添加参数量计算 def count_parameters(model): return sum(p.numel() for p in model.parameters() if p.requires_grad) total_params = count_parameters(yolact_net) print(f"Model Parameters: {total_params / 1e6:.3f}M") # 转换为百万单位 # 在模型初始化后添加 input_size = (1, 3, cfg.max_size, cfg.max_size) dummy_input = torch.zeros(*input_size).to("cuda" if args.cuda else "cpu") flops, _ = profile(yolact_net, inputs=(dummy_input,), verbose=False) print(f"GFLOPs: {flops / 1e9:.2f}G") if args.log: log = Log(cfg.name, args.log_folder, dict(args._get_kwargs()), overwrite=(args.resume is None), log_gpu_stats=args.log_gpu) # I don't use the timer during training (I use a different timing method). # Apparently there's a race condition with multiple GPUs, so disable it just to be safe. timer.disable_all() # Both of these can set args.resume to None, so do them before the check if args.resume == 'interrupt': args.resume = SavePath.get_interrupt(args.save_folder) elif args.resume == 'latest': args.resume = SavePath.get_latest(args.save_folder, cfg.name) if args.resume is not None: print('Resuming training, loading {}...'.format(args.resume)) yolact_net.load_weights(args.resume) if args.start_iter == -1: args.start_iter = SavePath.from_str(args.resume).iteration else: print('Initializing weights...') yolact_net.init_weights(backbone_path=args.save_folder + cfg.backbone.path) optimizer = optim.SGD(net.parameters(), lr=args.lr, momentum=args.momentum, weight_decay=args.decay) criterion = MultiBoxLoss(num_classes=cfg.num_classes, pos_threshold=cfg.positive_iou_threshold, neg_threshold=cfg.negative_iou_threshold, negpos_ratio=cfg.ohem_negpos_ratio) if args.batch_alloc is not None: args.batch_alloc = [int(x) for x in args.batch_alloc.split(',')] if sum(args.batch_alloc) != args.batch_size: print('Error: Batch allocation (%s) does not sum to batch size (%s).' % (args.batch_alloc, args.batch_size)) exit(-1) net = CustomDataParallel(NetLoss(net, criterion)) if args.cuda: net = net.cuda() # Initialize everything if not cfg.freeze_bn: yolact_net.freeze_bn() # Freeze bn so we don't kill our means yolact_net(torch.zeros(1, 3, cfg.max_size, cfg.max_size).cuda()) if not cfg.freeze_bn: yolact_net.freeze_bn(True) # loss counters loc_loss = 0 conf_loss = 0 iteration = max(args.start_iter, 0) last_time = time.time() epoch_size = len(dataset)+1 // args.batch_size num_epochs = math.ceil(cfg.max_iter / epoch_size) # Which learning rate adjustment step are we on? lr' = lr * gamma ^ step_index step_index = 0 data_loader = data.DataLoader(dataset, args.batch_size, num_workers=args.num_workers, shuffle=True, collate_fn=detection_collate, pin_memory=True) save_path = lambda epoch, iteration: SavePath(cfg.name, epoch, iteration).get_path(root=args.save_folder) time_avg = MovingAverage() global loss_types # Forms the print order loss_avgs = { k: MovingAverage(100) for k in loss_types } print('Begin training!') print() # try-except so you can use ctrl+c to save early and stop training try: for epoch in range(num_epochs): # Resume from start_iter if (epoch+1)*epoch_size < iteration: continue for datum in data_loader: # Stop if we've reached an epoch if we're resuming from start_iter if iteration == (epoch+1)*epoch_size: break # Stop at the configured number of iterations even if mid-epoch if iteration == cfg.max_iter: break # Change a config setting if we've reached the specified iteration changed = False for change in cfg.delayed_settings: if iteration >= change[0]: changed = True cfg.replace(change[1]) # Reset the loss averages because things might have changed for avg in loss_avgs: avg.reset() # If a config setting was changed, remove it from the list so we don't keep checking if changed: cfg.delayed_settings = [x for x in cfg.delayed_settings if x[0] > iteration] # Warm up by linearly interpolating the learning rate from some smaller value if cfg.lr_warmup_until > 0 and iteration <= cfg.lr_warmup_until: set_lr(optimizer, (args.lr - cfg.lr_warmup_init) * (iteration / cfg.lr_warmup_until) + cfg.lr_warmup_init) # Adjust the learning rate at the given iterations, but also if we resume from past that iteration while step_index < len(cfg.lr_steps) and iteration >= cfg.lr_steps[step_index]: step_index += 1 set_lr(optimizer, args.lr * (args.gamma ** step_index)) # Zero the grad to get ready to compute gradients optimizer.zero_grad() # Forward Pass + Compute loss at the same time (see CustomDataParallel and NetLoss) losses = net(datum) losses = { k: (v).mean() for k,v in losses.items() } # Mean here because Dataparallel loss = sum([losses[k] for k in losses]) # no_inf_mean removes some components from the loss, so make sure to backward through all of it # all_loss = sum([v.mean() for v in losses.values()]) # Backprop loss.backward() # Do this to free up vram even if loss is not finite if torch.isfinite(loss).item(): optimizer.step() # Add the loss to the moving average for bookkeeping for k in losses: loss_avgs[k].add(losses[k].item()) cur_time = time.time() elapsed = cur_time - last_time last_time = cur_time # Exclude graph setup from the timing information if iteration != args.start_iter: time_avg.add(elapsed) if iteration % 10 == 0: eta_str = str(datetime.timedelta(seconds=(cfg.max_iter-iteration) * time_avg.get_avg())).split('.')[0] total = sum([loss_avgs[k].get_avg() for k in losses]) loss_labels = sum([[k, loss_avgs[k].get_avg()] for k in loss_types if k in losses], []) print(('[%3d] %7d ||' + (' %s: %.3f |' * len(losses)) + ' T: %.3f || ETA: %s || timer: %.3f') % tuple([epoch, iteration] + loss_labels + [total, eta_str, elapsed]), flush=True) if args.log: precision = 5 loss_info = {k: round(losses[k].item(), precision) for k in losses} loss_info['T'] = round(loss.item(), precision) if args.log_gpu: log.log_gpu_stats = (iteration % 10 == 0) # nvidia-smi is sloooow log.log('train', loss=loss_info, epoch=epoch, iter=iteration, lr=round(cur_lr, 10), elapsed=elapsed) log.log_gpu_stats = args.log_gpu iteration += 1 if iteration % args.save_interval == 0 and iteration != args.start_iter: if args.keep_latest: latest = SavePath.get_latest(args.save_folder, cfg.name) print('Saving state, iter:', iteration) yolact_net.save_weights(save_path(epoch, iteration)) if args.keep_latest and latest is not None: if args.keep_latest_interval <= 0 or iteration % args.keep_latest_interval != args.save_interval: print('Deleting old save...') os.remove(latest) # This is done per epoch if args.validation_epoch > 0: if epoch % args.validation_epoch == 0 and epoch > 0: compute_validation_map(epoch, iteration, yolact_net, val_dataset, log if args.log else None) # Compute validation mAP after training is finished compute_validation_map(epoch, iteration, yolact_net, val_dataset, log if args.log else None) except KeyboardInterrupt: if args.interrupt: print('Stopping early. Saving network...') # Delete previous copy of the interrupted network so we don't spam the weights folder SavePath.remove_interrupt(args.save_folder) yolact_net.save_weights(save_path(epoch, repr(iteration) + '_interrupt')) exit() yolact_net.save_weights(save_path(epoch, iteration)) def set_lr(optimizer, new_lr): for param_group in optimizer.param_groups: param_group['lr'] = new_lr global cur_lr cur_lr = new_lr def gradinator(x): x.requires_grad = False return x def prepare_data(datum, devices:list=None, allocation:list=None): with torch.no_grad(): if devices is None: devices = ['cuda:0'] if args.cuda else ['cpu'] if allocation is None: allocation = [args.batch_size // len(devices)] * (len(devices) - 1) allocation.append(args.batch_size - sum(allocation)) # The rest might need more/less images, (targets, masks, num_crowds) = datum cur_idx = 0 for device, alloc in zip(devices, allocation): for _ in range(alloc): images[cur_idx] = gradinator(images[cur_idx].to(device)) targets[cur_idx] = gradinator(targets[cur_idx].to(device)) masks[cur_idx] = gradinator(masks[cur_idx].to(device)) cur_idx += 1 if cfg.preserve_aspect_ratio: # Choose a random size from the batch _, h, w = images[random.randint(0, len(images)-1)].size() for idx, (image, target, mask, num_crowd) in enumerate(zip(images, targets, masks, num_crowds)): images[idx], targets[idx], masks[idx], num_crowds[idx] \ = enforce_size(image, target, mask, num_crowd, w, h) cur_idx = 0 split_images, split_targets, split_masks, split_numcrowds \ = [[None for alloc in allocation] for _ in range(4)] for device_idx, alloc in enumerate(allocation): split_images[device_idx] = torch.stack(images[cur_idx:cur_idx+alloc], dim=0) split_targets[device_idx] = targets[cur_idx:cur_idx+alloc] split_masks[device_idx] = masks[cur_idx:cur_idx+alloc] split_numcrowds[device_idx] = num_crowds[cur_idx:cur_idx+alloc] cur_idx += alloc return split_images, split_targets, split_masks, split_numcrowds def no_inf_mean(x:torch.Tensor): """ Computes the mean of a vector, throwing out all inf values. If there are no non-inf values, this will return inf (i.e., just the normal mean). """ no_inf = [a for a in x if torch.isfinite(a)] if len(no_inf) > 0: return sum(no_inf) / len(no_inf) else: return x.mean() def compute_validation_loss(net, data_loader, criterion): global loss_types with torch.no_grad(): losses = {} # Don't switch to eval mode because we want to get losses iterations = 0 for datum in data_loader: images, targets, masks, num_crowds = prepare_data(datum) out = net(images) wrapper = ScatterWrapper(targets, masks, num_crowds) _losses = criterion(out, wrapper, wrapper.make_mask()) for k, v in _losses.items(): v = v.mean().item() if k in losses: losses[k] += v else: losses[k] = v iterations += 1 if args.validation_size <= iterations * args.batch_size: break for k in losses: losses[k] /= iterations loss_labels = sum([[k, losses[k]] for k in loss_types if k in losses], []) print(('Validation ||' + (' %s: %.3f |' * len(losses)) + ')') % tuple(loss_labels), flush=True) # 修改 compute_validation_map 函数 def compute_validation_map(epoch, iteration, yolact_net, dataset, log: Log = None): with torch.no_grad(): yolact_net.eval() # 添加 FPS 计算 num_test_frames = 100 total_time = 0 # 预热 GPU for _ in range(10): _ = yolact_net(torch.zeros(1, 3, cfg.max_size, cfg.max_size).cuda()) # 正式测试 for i in range(num_test_frames): img, _ = dataset[i] img = img.unsqueeze(0).cuda() start_time = time.perf_counter() preds = yolact_net(img) torch.cuda.synchronize() # 确保 CUDA 操作完成 total_time += time.perf_counter() - start_time fps = num_test_frames / total_time print(f"FPS: {fps:.2f}") # 原有验证代码 print("\nComputing validation mAP...") val_info = eval_script.evaluate(yolact_net, dataset, train_mode=True) # 记录 FPS if log is not None: log.log('val', {'fps': fps}, epoch=epoch, iter=iteration) yolact_net.train() return fps # 在 compute_validation_map 函数中 print(f"\nValidation Metrics @ iter {iteration}:") print(f"├── Params: {total_params / 1e6:.3f}M") print(f"├── GFLOPs: {flops / 1e9:.2f}G") print(f"├── FPS: {fps:.2f}") print(f"└── mIoU: {val_info.get('mIoU', 0):.4f}") # 记录所有指标 if log is not None: metrics = { 'params': total_params, 'gflops': flops / 1e9, 'fps': fps, 'mIoU': val_info.get('mIoU', 0) } log.log('metrics', metrics, epoch=epoch, iter=iteration) def setup_eval(): eval_script.parse_args(['--no_bar', '--max_images='+str(args.validation_size)]) if __name__ == '__main__': train()

我们正在修改YOLACT的训练脚本(train.py)，目的是在模型初始化后添加FLOPs和参数量的计算。根据用户提供的错误信息，问题发生在使用thop.profile计算FLOPs时，因为模型已经被ScriptModule化（可能由于某些原因已经被...

解释一下以下代码：def train_model(model, criterion, optimizer, scheduler, num_epochs=25): since = time.time() # 返回当前时间的时间戳 # deepcopy为深拷贝,即创建一个新的对象，完全复制原始对象及其所有子对象。 # model.state_dict是PyTorch中的一个方法，返回一个字典对象，将模型的所有参数映射到它们的张量值 best_model_wts = copy.deepcopy(model.state_dict()) best_acc = 0.0 # 显示当前的训练进度 for epoch in range(num_epochs): print('Epoch {}/{}'.format(epoch, num_epochs - 1)) # 输出训练进度 print('-' * 10) # 每个epoch都有一个训练和验证阶段 for phase in ['train', 'val']: if phase == 'train': # 调用Optimizer的step函数使它所有参数更新 optimizer.step() # PyTorch中用于更新优化器的学习率 scheduler.step() model.train() # 设置模型的工作模式：训练模式，模型会更新参数 else: model.eval() # 设置模型的工作模式：评估模式，模型不会更新参数 running_loss = 0.0 # 用于记录当前epoch的累计损失 running_corrects = 0 # 用于记录当前正确分类的数量 # 迭代数据 for inputs, labels in dataloaders[phase]: # GPU加速,inputs、labels用于将输入数据和标签数据移动到指定的设备上，例如CPU或GPU inputs = inputs.to(device) labels = labels.to(device) # 清空梯度，在每次优化前都需要进行此操作 optimizer.zero_grad() # torch.set_grad_enabled在训练模式下启用梯度计算，而在评估模式下禁用梯度计算 with torch.set_grad_enabled(phase == 'train'): outputs = model(inputs) _, preds = torch.max(outputs, 1) # 返回输入张量在指定维度上的最大值及其对应的索引 loss = criterion(outputs, labels) # 后向 + 仅在训练阶段进行优化 if phase == 'train': # 反向传播：根据模型的参数计算loss的梯度 loss.backward() # 调用Optimizer的step函数使它所有参数更新 optimizer.step() running_loss += loss.item() * inputs.size(0) # 获取损失值和输入数据的数量 running_corrects += torch.sum(preds == labels.data) # 记录当前epoch中正确分类的数量 # dataset_sizes[phase]：该变量用于获取数据集在当前阶段中的大小 epoch_loss = running_loss / dataset_sizes[phase] #当前epoch的平均损失 epoch_acc = running_corrects.double() / dataset_sizes[phase] # 当前epoch的准确率 # 打印当前epoch的损失和准确率，以实时监控模型的训练状态 print('{} Loss: {:.4f} Acc: {:.4f}'.format( phase, epoch_loss, epoch_acc)) # 前阶段的名称、当前epoch的平均损失、当前epoch的准确率 if phase == 'val' and epoch_acc > best_acc: #更新最佳模型参数 best_acc = epoch_acc best_model_wts = copy.deepcopy(model.state_dict()) print() time_elapsed = time.time() - since # 输出整个训练过程的时间 print('Training complete in {:.0f}m {:.0f}s'.format( time_elapsed // 60, time_elapsed % 60)) # 用于输出最佳验证准确率,best_acc是训练过程中被更新为模型在验证集上的最佳准确率 print('Best val Acc: {:4f}'.format(best_acc)) # 用于加载模型的状态字典，即将最佳模型的参数设置为给定的参数字典 model.load_state_dict(best_model_wts) return model

模型参数保存（Model Checkpoint Saving） 下面将逐一解释这些部分。 ### 1. 训练循环（Training Loop） 训练循环是模型训练的核心，通常在每个epoch中遍历整个训练数据集。主要步骤包括： - 将模型设置为...
zip

基于Django的酒店预订信息管理系统

酒店预订信息管理系统是基于Django框架开发的，用于提升酒店业和旅游业的效率和用户体验。系统采用Django的MTV架构模式，实现模块化开发，便于维护和团队协作。系统包含用户注册登录、酒店信息展示、在线预订、订单管理、用户评论反馈等核心功能。用户注册登录模块支持信息的新增、编辑、查询和删除，并采用密码加密技术保障数据安全。酒店信息展示模块通过数据库存储和管理酒店的详细信息，如房间类型、价格、设施和图片。在线预订模块设计预订流程，包括房间选择、时间设定和支付方式，确保用户便捷完成预订。订单管理模块供酒店管理者查看和处理订单，具备权限管理功能。用户评论反馈模块为用户和酒店提供互动平台，用户可评价服务，酒店可回复反馈。Django框架提供数据库支持、用户认证系统、内容管理系统等组件，提高开发效率和系统稳定性。系统可部署在多种环境中，支持传统服务器和Docker容器化部署。该系统为酒店提供全面的数字化解决方案，提升管理效率和用户体验，是酒店业数字化转型的重要工具。资源来源于网络分享，仅用于学习交流使用，请勿用于商业，如有侵权请联系我删除！
doc

Selenium实现自动登录脚本.doc

Selenium实现自动登录脚本.doc
doc

滑动窗口算法模板及刷题应用场景详解.doc

滑动窗口算法模板及刷题应用场景详解.doc
rar

美格SMR815系列资料.rar

美格SMR815系列资料
doc

TDA3000+EOG招标参数_CN_2012-2H.doc

TDA3000+EOG招标参数_CN_2012-2H.doc
doc

Vue项目中封装通用Dialog弹窗组件.doc

Vue项目中封装通用Dialog弹窗组件.doc
doc

系统集成智能化工程专业技术方案.doc

系统集成智能化工程专业技术方案.doc
doc

Redis持久化机制RDB与AOF全解析.doc

Redis持久化机制RDB与AOF全解析.doc
zip

实现多个设备同时充电的示例效果.zip

1.版本：matlab2014a/2019b/2024b 2.附赠案例数据可直接运行。 3.代码特点：参数化编程、参数可方便更改、代码编程思路清晰、注释明细。 4.适用对象：计算机，电子信息工程、数学等专业的大学生课程设计、期末大作业和毕业设计。
docx

【Python编程】基于Python的网络爬虫技术详解：数据采集与处理的基本流程与实践指南

内容概要：本文介绍了Python爬虫的基础知识、常见步骤及注意事项。首先概述了Python爬虫的概念，即一种用Python编写的程序，可自动从网络获取并提取网页信息，广泛应用于数据采集、信息监测等领域。接着详细阐述了爬虫的五个常见步骤：确定目标与分析网页，明确所需数据及其呈现形式；发送请求，利用urllib或requests库获取网页内容；解析网页内容，借助BeautifulSoup或lxml库提取有用信息；提取数据，依据网页结构定位并获取目标数据；数据存储，将提取的数据保存为CSV、JSON或存入数据库。最后强调了使用爬虫时应注意遵守法律法规、遵循网站规则、避免过度请求等事项。; 适合人群：对Python爬虫感兴趣的初学者，包括但不限于程序员、数据分析师等。; 使用场景及目标：①学习Python爬虫的基本概念与工作原理；②掌握构建简单爬虫程序的技能，能够独立完成从确定目标到数据存储的全过程。; 阅读建议：由于涉及到具体的代码示例，在阅读过程中应结合实际操作练习，同时注意遵守文中提到的法律法规与道德规范，确保爬虫活动合法合规。
pdf

2025年人工智能AI指数报告.pdf

2025年人工智能AI指数报告.pdf
zip

重新构建的Dao Jet数据库引擎

资源下载链接为： https://pan.quark.cn/s/9e7ef05254f8 DAO Jet 数据库引擎是微软早期开发的用于操作 Access 数据库的组件，与 Jet 引擎紧密相连。Jet 引擎是 Windows 操作系统内置的数据库管理系统，主要支持 Access 数据库及其他使用 Jet 数据存储格式的应用程序。DAO（数据访问对象）是与 Jet 引擎搭配的数据访问接口，能让程序员利用对象和方法操作数据库。当出现“无法初始化 DAO/Jet 数据库引擎”的问题时，通常是由于 DAO 或 Jet 引擎组件损坏、丢失或版本不兼容引起的，这可能是系统更新、病毒入侵、卸载不当或其他软件冲突所致。解决办法之一是从 VC6 光盘中提取相关文件重新安装 DAO Jet 数据库引擎。DAO 接口包含多种类和接口，如 Database、Recordset、Field 等，开发者可借助这些对象执行 SQL 查询、创建和修改表、索引及查询等操作。与 ADO（ActiveX 数据对象）相比，DAO 功能稍弱，但因更贴近底层的 Jet 引擎，在处理 Access 特定特性时可能更便捷。安装 DAO Jet 数据库引擎一般包括以下步骤：下载包含所有必要 DAO 组件的 DAO-REDIST 文件；运行安装程序，按向导提示操作；安装时系统会自动检测并修复 DAO 组件问题或安装缺失组件；安装完成后，可能需重启计算机使更改生效；之后可通过编写简单 VBA 代码测试数据库连接，验证 DAO 是否正常工作。需注意，DAO Jet 数据库引擎主要适用于旧系统和应用程序，新版本的 Microsoft Office 和 Windows 操作系统不再推荐使用 DAO，而是更倾向于采用 ADO 接口以及 SQL Server Express 等更安全、更强大的数据库管理系统。不过，对于依赖 DAO 技
docx

【AI编程与智能应用开发】基于扣子平台COZE的编程学习助手智能体构建与实战案例分析

内容概要：文章详细介绍了字节跳动推出的智能体开发平台——扣子（Coze），重点讲解了如何在该平台上构建一个“编程学习助手”智能体。此智能体能够支持自然语言编程提问、展示代码示例、集成 GitHub 插件查询开源项目、支持语义上下文记忆，并可部署至微信、网页等多个端口。通过图形化流程设计、系统提示词引导、插件功能扩展以及上下文记忆等功能，Coze 平台大大降低了 AI 应用开发的门槛，使得非程序员也能创建实用的智能产品。此外，Coze 还提供了便捷的跨平台部署方式，增强了智能体的实用性和分发便利性。; 适合人群：对 AI 编程和智能应用开发感兴趣的初学者及个人开发者，特别是那些希望快速上手但缺乏传统编程经验的人群。; 使用场景及目标：①利用图形化界面和提示词创建智能体，降低开发门槛；②构建具备自然语言处理能力的编程学习助手，提高编程学习效率；③通过集成 GitHub 插件，查询和推荐开源项目资源；④实现智能体的多平台部署，扩大应用范围。; 其他说明：Coze 平台不仅适用于个人开发者进行实验性项目，还被广泛应用于企业客服、教育、运营自动化等领域。随着插件生态和大模型技术的发展，Coze 将在未来继续拓展其应用场景和开发深度。

看一下修改时间：Image$$VECTOR$$Base

Image\$\$VECTOR\$\$Base Image$$VECTOR$$Base 客服测试
pdf

2025年我国制造业数字化转型发展形势展望.pdf

2025年我国制造业数字化转型发展形势展望.pdf

大家在看

recommend-type

ray-optics:光学系统的几何光线追踪

射线光学 安装 要使用pip安装rayoptics ，请使用 > pip install rayoptics 或者，可以使用conda从conda - forge渠道安装rayoptics > conda install rayoptics --channel conda-forge 文献资料 射线光学位于“ 成像光学设计和分析工具 RayOptics是一个Python几何光学和成像光学库。 它为分析成像和相干光学系统提供了几何射线追踪基础。 在此基础上提供了许多标准的几何分析选项，例如横向射线和波前像差分析。 y-ybar图和镜头布局视图中近轴光线的图形编辑也支持光学系统的近轴布局。 支持导入Zemax .zmx和CODEV .seq文件。 RayOptics可用于Python脚本，Python和IPython外壳，Jupyter笔记本以及基于Qt的图形用户界面应用程序中。 笔记 该项
recommend-type

修复Windows 10&11 因更新造成的IE11 无法使用

修复Windows 10&11 因更新造成的IE11 无法使用
recommend-type

参考资料-Boost_PFC电路中开关器件的损耗分析与计算.zip

参考资料-Boost_PFC电路中开关器件的损耗分析与计算.zip
recommend-type

3DSlicer 5.2带中文包-稳定版

这是官方2023-02月发布发布的稳定版3DSlicer，里面已经安装了常用插件，并且做了分类处理，常用工具放在了智能医学所属栏里面，附带了中文包，可直接在设置里面选择中文。本软件适用于医学影像处理初学者。可根据用户习惯添加或者删除模块。
recommend-type

KGM转MP3或者FLAC_kgma_kgma格式_FLAC_kgma转换器_kgm转换成flac_亲测完美转换！保证可用。

使用方法：直接将带转换的KGM文件或者KGMA文件放在文件夹里，（可批量放置），将kgm音乐文件复制到"KGM转MP3或者FLAC"文件夹内，运行unlock-kugou-windows-amd64-alpha2.exe，等待转换完成即可。最终会输出未加密的MP3文件或者FLAC文件，使用任何播放器均可直接打开。最终输出的文件会在kgm-vpr-out文件夹中

最新推荐

recommend-type

PKID查壳工具最新版发布，轻松识别安卓安装包加壳

根据提供的文件信息，我们可以详细解读以下知识点： ### PKiD(查壳)工具介绍 #### 标题分析 - **PKiD(查壳)**: 这是一个专门用于分析安卓安装包（APK文件）是否被加壳的应用程序。"查壳"是一种用于检测软件是否被保护层（即“壳”）包裹的技术术语。加壳是一种常见的软件保护手段，用于隐藏真实的代码逻辑，防止恶意逆向分析。 - **RAR格式文件**: 文件使用了RAR格式进行压缩，这是WinRAR软件用于文件压缩和解压缩的专有格式。 #### 描述分析 - **ApkScan-PKID查壳工具.zip**: 这指的是一款名为ApkScan的工具，它包含了PKID查壳功能。该工具被打包成ZIP格式，便于用户下载和使用。 - **安卓安装包**: 这是指Android平台的应用程序安装包，通常以APK作为文件扩展名。 - **加壳检测**: PKID查壳工具用于检测APK文件是否被加壳，加壳是一种常见的软件保护技术，用于加密和保护软件免遭逆向工程。 - **脱壳测试**: 如果检测到加壳，脱壳测试将用于尝试去除或绕过保护层，以便进行安全分析、调试或修改程序。 #### 标签分析 - **查壳**: 再次强调了工具的主要功能，即检测APK文件中的加壳情况。 - **最新版**: 表示这个文件是PKID查壳工具的最新版本。 - **PKID**: 这是工具的核心名称，代表着该软件的主要功能和用途。 #### 文件列表分析 - **PKiD(查壳).exe**: 这是一个可执行文件，说明PKID查壳工具是一个独立的应用程序，用户可以通过双击此文件直接运行程序，而无需安装。 ### 技术背景 #### 查壳工具的工作原理 查壳工具通常通过分析APK文件的头部信息、资源文件和代码段来检测加壳。它可能会检查PE文件格式的特定区域（APK基于DEX，但PE检查的概念相似），这些区域在加壳过程中可能会被特定的代码模式、字符串或签名标记。例如，某些壳会在文件头部加入特定的字符串，或者修改方法计数等信息。 #### 加壳技术 加壳技术通常用于防止软件被轻易反编译或逆向工程。它可以阻止潜在的窃取知识产权、绕过付费或防止代码分析等。加壳过程包括加密和压缩原始代码，然后在运行时解压和解密代码以供执行。 #### 脱壳技术 脱壳技术是指绕过或移除软件保护壳的过程。这通常需要对壳的工作原理有深入了解，并且需要一定的逆向工程技能。脱壳过程可能包括识别壳的签名、分析保护机制、修改代码和重定位等步骤。 #### 安全测试与逆向工程 查壳工具对于安全测试和逆向工程人员来说是非常重要的工具。它们可以帮助识别软件是否被加壳，从而决定是否需要进行脱壳操作以便深入分析软件的安全漏洞或版权问题。 #### 使用场景 - **安全测试**: 安全工程师在对移动应用程序进行安全评估时，需要确认是否有必要进行脱壳处理。 - **逆向工程**: 开发人员可能需要逆向工程某个应用，以更好地理解其工作原理或为了兼容性问题进行修改。 - **软件保护**: 对于软件开发者而言，了解常见的加壳和脱壳技术有助于更好地保护自己的应用免遭未授权的篡改和分析。 ### 结论 了解和掌握PKID查壳工具的使用对于任何需要对APK文件进行安全性分析的专业人士来说都是基础且关键的技能。该工具提供了一种快速判断软件是否加壳的方法，而且在发现壳的情况下，还可以推测出壳的来源公司，这对于后续的分析和处理提供了便利。然而，值得注意的是，进行脱壳操作往往需要对原始程序代码进行逆向工程，这在很多情况下可能涉及法律和道德问题，因此在使用此类工具时，必须确保其合法性和符合道德规范。
recommend-type

【PDF.js问题诊断手册】：快速定位与解决常见问题

# 摘要 PDF.js是一个广泛应用于Web环境中的PDF文件渲染库，其基本原理和应用概况将在本文第一章进行概述。第二章深入探讨PDF.js的问题诊断基础，包括架构分析、问题分类和日志分析技巧。随后，本文将具体讨论PDF.js在显示问题、加载与性能问题以及兼容性与安全问题的诊断与
recommend-type

grep -Ei

<think>首先，用户的问题是关于使用grep命令进行忽略大小写的扩展正则表达式搜索。用户指定了："grep -Ei ignore case and use extended regular expressions"。所以，我需要解释如何使用grep的-E和-i选项。 从系统指令中： - 所有行内数学表达式必须用$...$格式。 - 独立公式用$$...$$。 - 使用中文回答。 - 生成相关问题。 - 在回答中引用的段落末尾添加引用标识，如[^1]。 - 回答结构清晰，逐步解决问题。 参考引用： - 引用[1]提到使用-E选项进行扩展正则表达式，而不是基本正则表达式。这更清晰，因为反斜
recommend-type

一键关闭系统更新的工具介绍

从给定的文件信息中我们可以分析出几个相关的知识点，以下是详细说明： 【标题】“系统禁止更新工具.7z”暗示着这个压缩文件内包含的可能是一款软件工具，其主要功能是阻止或禁止操作系统的更新。这种工具可能针对的是Windows、Linux或者其他操作系统的自动更新功能。一般来说，用户可能出于稳定性考虑，希望控制更新时间，或者是因为特定的软件环境依赖于旧版本的系统兼容性，不希望系统自动更新导致兼容性问题。 【描述】“一健关闭系统更新”说明了该工具的使用方式非常简单直接。用户只需通过简单的操作，比如点击一个按钮或者执行一个命令，就能实现关闭系统自动更新的目的。这种一键式操作符合用户追求的易用性原则，使得不太精通系统操作的用户也能轻松控制更新设置。 【标签】“系统工具”表明这是一个与操作系统紧密相关的辅助工具。系统工具通常包括系统清理、性能优化、磁盘管理等多种功能，而本工具专注于管理系统更新，使其成为系统维护中的一环。 【压缩包子文件的文件名称列表】“系统禁止更新工具”是压缩包内的文件名。由于文件格式为“.7z”，这说明该工具采用了7-Zip压缩格式。7-Zip是一款开源且免费的压缩软件，支持非常高的压缩比，并且能够处理各种压缩文件格式，如ZIP、RAR等。它支持创建密码保护的压缩文件和分卷压缩，这在需要转移大量数据时特别有用。然而在这个上下文中，“系统禁止更新工具”文件名暗示了该压缩包内只包含了一个程序，即专门用于关闭系统更新的工具。 根据标题和描述，我们可以推测该工具可能的实现机制，例如： 1. 修改系统服务的配置：在Windows系统中，可以通过修改Windows Update服务的属性来禁用该服务，从而阻止系统自动下载和安装更新。 2. 修改注册表设置：通过编辑Windows注册表中的某些特定键值，可以关闭系统更新功能。这通常涉及到对HKEY_LOCAL_MACHINE\SOFTWARE\Policies\Microsoft\Windows\WindowsUpdate和HKEY_LOCAL_MACHINE\SOFTWARE\Policies\Microsoft\Windows\WindowsUpdate\AU等路径下的设置进行修改。 3. 使用第三方软件：某些第三方工具提供了一键禁用系统更新的功能，通过修改操作系统的相关文件和配置来实现更新的禁用。这包括阻止Windows Update的运行、拦截更新下载等。 该工具的使用可能带来一些后果，比如系统安全风险的增加、系统漏洞得不到及时修复以及可能违反某些软件许可协议。用户在使用这类工具时应该清楚这些潜在的影响，并自行承担相应的风险。 总结来说，该“系统禁止更新工具”可以视为一个针对特定用户需求，尤其是企业用户或个人用户中对系统稳定性有较高要求的人群，提供的一个便捷的解决方案。它通过直接修改系统设置来禁止操作系统自动更新，但用户在使用此工具时需要谨慎，因为这可能会带来系统安全和稳定性的风险。
recommend-type

【PC3000高级应用】：复杂故障快速诊断与解决

# 摘要 本文详细介绍了PC3000在数据恢复领域的应用及其重要性，阐述了其基本操作、界面理解以及在复杂故障诊断中的应用方法。深入剖析了PC3
recommend-type

我的数据是names(data) [1] "分组" "排序" "PACU" "V1" "V2" "V3" ；后四个数据分别代表四个时间点的测量值，请你用ART模型进行规范检验，我需要得到分组（即不同处理方法）的差异

<think>我们正在处理一个重复测量设计的数据分析问题，用户希望使用ART（Align Rank Transform）方法分析包含分组因素和多个时间点的数据。数据格式包括分组变量（group）、受试者编号（subject）以及四个时间点（V1-V3）的测量值。 ART分析步骤： 1. 数据准备：将宽格式数据转换为长格式（每个时间点一行） 2. 使用ARTool包进行对齐秩变换 3. 构建混合效应模型（包含分组、时间及其交互） 4. 对变换后的数据进行方差分析 5. 进行事后检验（如存在显著效应） 引用说明： - 用户提供的引用[1][2][3][4]与当前问题无关，因此忽略这些引用 -
recommend-type

鼎捷易飞ERPV9.0委外进货单批量导入解决方案

根据提供的文件信息，我们可以从标题、描述、标签以及压缩包文件列表中提取以下知识点： 1. 委外进货单批量导入程序及模版格式 标题提到的“委外进货单批量导入程序”指的是一个软件应用，其主要功能是允许用户批量地将委外进货数据导入到ERP系统中。批量导入通常是指在ERP系统中不需要逐条手动输入数据，而是通过预先定义好的模板，一次性将大量数据导入系统。这样的程序对于提高工作效率、减少重复性工作以及避免人为错误具有重要意义。 2. 鼎捷易飞ERPV9.0 描述中提到的“鼎捷易飞ERPV9.0”是一个特定版本的ERP系统，由鼎捷软件公司开发。ERP（Enterprise Resource Planning，企业资源计划）系统是一种用于整合企业内部所有资源信息，实现信息流、物流、资金流、工作流的高度集成和自动化管理的软件。ERPV9.0是该系列产品的版本号，表明该程序和文件模板是为这一特定版本的ERP系统设计。 3. .NET C#源代码 标题中的“.NET C#源代码”表示程序是使用.NET框架和C#语言开发的。.NET是微软公司开发的一个软件框架，用于构建和运行Windows应用程序。C#（读作“C Sharp”）是.NET框架下的一种编程语言，具有面向对象、类型安全和垃圾回收等特点。开发者可能提供了源代码，以便企业用户可以自行修改、调整以满足特定需求。 4. 使用方法和步骤 描述中详细说明了程序的使用方法： - 首先编辑模版格式数据，即将需要导入的数据按照特定的格式要求填写到模板中。 - 然后在程序中选择单别（可能指的是单据类型）和日期等条件。 - 点击“导入数据”按钮，程序将提示用户选择含有数据的模板文件。 - 程序会进行数据校验，以确保数据的正确性。校验规则是特定的，如果用户不需要特定的校验条件，可以在程序中直接删除这部分代码。 - 最后，数据校验无误后，程序可以生成相应的进货单据。 5. 自定义程序和模板 在标签中提到的“易飞ERP委外进货单导入程序”、“委外进货单导入程序”和“易飞ERP自定义程序”表明，所提供的程序可以根据用户的特定需求进行定制。同时，模板格式的使用，也意味着用户可以根据自己的业务需要和ERP系统的要求调整模板内容，以便更好地适应数据导入过程。 6. 压缩包子文件的文件名称列表 提供的压缩包包含两个文件： - “委外进货单批量导入格式.xls”指的是一个Excel格式的文件，这个文件应该是一个符合要求的导入模板，用户需要按照这个模板的格式来准备需要导入的数据。 - “委外入库导入(源代码).zip”是一个包含源代码的压缩文件，其中可能包含程序的完整源代码，也可能仅包含与导入功能相关的部分代码。 7. 文件列表和数据导入流程 通过文件列表可以理解整个数据导入流程： - 用户首先需要准备好数据，并将其按照“委外进货单批量导入格式.xls”模板中的要求填入相应的数据。 - 用户然后使用“委外入库导入(源代码).zip”中的程序，根据程序使用说明进行配置。 - 最后，用户通过程序界面导入整理好的Excel模板文件，程序会自动处理数据导入的过程，并进行必要的数据校验，最终生成所需的进货单据。 以上总结的知识点涵盖了该压缩包文件的内容、用途、技术细节以及操作步骤，为了解和使用该委外进货单批量导入程序提供了充分的信息。
recommend-type

硬盘数据恢复秘籍：PC3000流程全面解读

# 摘要 本文全面概述了硬盘数据恢复的各个方面，从基础理论到实际操作，再到行业规范和面临的挑战。首先介绍了硬盘数据恢复的基本概念和PC3000工具的工作原理，随后详细讨论了PC3000的硬件组成、软件功能及操作流程。接着，深入探讨了硬盘的基础知识、常见故障类型、故障识别方法以及数据恢复前的准备工作。通过实际案例演练，文章展示了PC3000在模拟故障和实际数据恢复中的应用，同时强调了高级数据恢复技巧
recommend-type

MRAM在PLC中起到的作用

<think>首先，用户的问题是基于之前的引用内容，特别是引用[1]，它讨论了MRAM在PLC中的应用。用户明确说：“我想了解MRAM在PLC中的作用”，并问“MRAM在可编程逻辑控制器（PLC）中的功能和应用”。 关键引用是引用[1]： - MRAM具有纳秒级高速写入、万亿次擦写寿命和宽工作温区。 - PLC是工业自动化系统的核心，需要快速读写和高擦写寿命。 - 传统方案使用SRAM和电池，但存在系统复杂、硬件成本高、数据丢失风险。 - MRAM可以替代这种方案，因为它支持连续20年毫秒级间隔的擦写操作。 - 在异常掉电时，MRAM能快速保存数据，无需备份电池。 - 具体例子：数字输入模块
recommend-type

局域网内FTP服务安装与搭建指南

在局域网中搭建FTP服务是网络管理和文件共享的常见需求，它允许用户在一个内部网络上创建一个集中式的文件存储和交换点。以下是关于局域网中FTP服务安装和搭建的知识点，将从FTP服务的概念、安装和配置等方面进行详细解释。 ### 1. FTP服务概念 **FTP（File Transfer Protocol）** 是文件传输协议的缩写，它是一种用于在网络上进行文件传输的标准协议。在局域网中搭建FTP服务，实质上是在网络中的某台服务器上安装FTP服务软件，使得局域网内的其他用户可以通过FTP客户端软件连接到这台服务器，并实现文件的上传和下载。 ### 2. FTP服务组件 一个标准的FTP服务系统一般由以下组件组成： - **FTP服务器**：负责响应客户端的请求，管理用户认证，以及文件的上传和下载。 - **FTP客户端**：用户用来连接服务器并进行文件传输的软件工具。 - **用户账户**：用于验证连接FTP服务器的用户身份，可设定不同的权限以管理对文件的访问。 - **目录结构**：服务器上存储文件的组织形式，通常根据权限和使用目的进行分级和分区。 ### 3. FTP服务软件选择 在局域网中搭建FTP服务，有多种软件可供选择，包括但不限于： - **FileZilla Server**：开源且功能强大的FTP服务器软件。 - **Serv-U**：一款功能丰富的商业FTP服务器软件，易于配置和管理。 - **Windows FTP Server**：适用于Windows系统的简单易用的FTP服务器。 ### 4. FTP服务搭建步骤 1. **安装FTP服务器软件**： 以FileZilla Server为例，首先需要从官方网站下载安装包。解压缩后执行安装程序，按照提示完成安装。 2. **配置FTP服务器软件**： 安装完成后，通常会有配置向导引导进行初步设置，包括监听的端口（默认为21），是否允许匿名访问，以及创建用户账户等。 3. **用户账户设置**： 根据需要创建用户账户，并分配合适的权限，如访问、读取、写入、删除等，来管理用户对FTP服务器上文件的操作。 4. **目录和权限设置**： 设定FTP服务器上的目录结构，设置每个目录的权限，确保用户只能访问到他们被允许的区域。 5. **启动FTP服务**： 配置完成后，启动FTP服务器服务，使之开始监听客户端的连接请求。 6. **测试FTP服务**： 使用FTP客户端软件（例如FileZilla Client）进行连接测试，验证安装和配置的正确性，并检查文件的上传下载是否可以正常工作。 ### 5. 安全和维护 在运行FTP服务的过程中，需要关注安全性和服务维护： - **安全性**：实施安全措施，如使用SSL/TLS加密传输，设置防火墙规则，以及定期更新FTP服务器软件。 - **维护**：定期检查服务器日志，监控异常活动，及时更新软件以修补安全漏洞。 ### 6. FTPbdfwkf.exe文件分析 关于“FTPbdfwkf.exe”，从文件名上看，这可能是一个特定的FTP服务器软件的可执行文件。但此名称并不对应常见的FTP软件。在没有具体上下文的情况下，很难确定这个文件的确切功能。不过，对于任何未知的可执行文件，重要的是通过安全软件进行扫描，以确保其不是恶意软件，并且来自于可信的来源。 在安装任何FTP服务软件之前，建议先对软件的来源和用户评价进行详细的了解，确保所选的FTP服务软件能够满足局域网文件共享的需求，且具备良好的兼容性和安全性。