Implementation & Design Review

Under the Esteemed Guidance Of

RETINAL FUNDUS
Mrs. Ch. DISEASES
Bala Vijaya DETECTION
Durga M.Tech, (Ph. D.)
Associate Professor

Submitted by IV CSM
(Batch-08)
Ms. SH AI K KUL S UM (21KT1A4256)
Ms. C H I GU R UPAT I SO W MYA (21KT1A4206)
Mr. MANU KONDA PR A NE E T H KU MAR (21KT1A4238)
 CONVERSION OF
2D BLUEPRINTS INTO
3D MODEL USING
NEURAL NETWORKS
ABSTRACT
 ABSTRACT
• Focuses on automated conversion of 2D blueprints into 3D models using
neural networks.
• Utilizes image processing techniques and deep learning algorithms to
analyze and interpret 2D blueprint drawings.
• Extracts key features such as dimensions, shapes, and annotations from
2D drawings.
• Processes the extracted features with a neural network trained to
reconstruct the corresponding 3D structure.
• Ensures efficient and accurate generation of 3D models, reducing manual
effort in traditional design workflows.
• Offers a scalable and robust solution for 3D reconstruction tasks.
INTRODUCTION
 INTRODUCTION
• Managing 2D blueprints is challenging and time-intensive.

• Converting 2D to 3D models requires specialized skills and

effort.

• Neural networks offer solutions with advanced image

recognition.

• Automation improves accuracy, saves time, and reduces effort.

• Supports researchers and professionals in handling complex

designs.
LITERATURE
SURVEY
 LITERATURE SURVEY
S.NO Author Title Dataset Classification Merits Demerits Accuracy
Algorithm

1 Jia Zheng, Deep Line GANs Combines Limited by the 91%

Learning drawings, deep complexity of
Yifan Zhu
Assisted CAD learning 2D line
Optimization models with drawings
for 3D geometric
Reconstructio constraints
n from Single to improve
2D Line reconstructi
Drawings on
accuracy
 LITERATURE SURVEY
S.NO Author Title Dataset Classification Merits Demerits Accuracy
Algorithm

2 Chen, Automatic Architect Radial Basis Accurate Computational 90%

Reconstructio ural Function 3D model ly intensive for
Sofia
n of 3D blueprints Networks reconstructi complex
Fiest,
Models from and CAD on from models
Luis 2D Drawings data diverse 2D
Jacques drawings
de Souza
 LITERATURE SURVEY
S.NO Author Title Dataset Classification Merits Demerits Accuracy
Algorithm

3 Sungsoo 3DPlanNet: Floor Ensemble Improves Limited to 89%

Park, Generating plans, methods, Multi- architectur structured
3D Models annotate view networks al floor plans
Kim
from 2D Floor d building visualizatio
Plan Images layouts n
 LITERATURE SURVEY
S.NO Author Title Dataset Classification Merits Demerits Accuracy
Algorithm

4 Patel, Conversion of Engineeri Variational Simplifies Difficulty in 88%

2D Blueprints ng Autoencoders deformable handling
Nasreen,
into 3D blueprints object highly
Shezan, Models , CAD reconstructi complex
Niha, data on deformable
objects
Preksha,
Shshira
 LITERATURE SURVEY
S.NO Author Title Dataset Classification Merits Demerits Accuracy
Algorithm

5 Robin BluNF: CAD and Neural fields Provides Requires high 88%
Courant, Blueprint architectu intuitive 3D computational
Neural Field ral scene resources
Xi Wang,
drawings editing
March tools
Christie
 LITERATURE SURVEY
S.NO Author Title Dataset Classification Merits Demerits Accuracy
Algorithm

6 J. Ryan 3D Neural 3D Triplane diffusion, Efficiently High memory 87%

Shue, Field training Neural fields generates requirements
Generation scenes 3D models during training
Ryan Po,
using Triplane with depth
Ankner, Diffusion accuracy
Jiajun,
Gorden
 LITERATURE SURVEY
S.NO Author Title Dataset Classification Merits Demerits Accuracy
Algorithm

7 Shoaib Investigating 3D object ANN Provides Limited to 86%

Ahmed the Nature of datasets insights into supervised
Siddiqui, 3D 3D classification
Generalizatio generalizati tasks
David,
n in Deep on
Thomas Neural capabilities
Networks of neural
networks
 LITERATURE SURVEY
S.NO Author Title Dataset Classification Merits Demerits Accuracy
Algorithm

8 Hamed New 2D to 3D Single 2D Perceptron Accurate Computational 86%

Amiri, Reconstructio line 3D complexity
Hannah n from Single drawings, reconstructi increases with
Vogel 2D Line Public ons from more complex
Drawings CAD single-line designs. May
Oliver
dataset, 2D fail with
synthetic drawings, intricate 3D
data efficient for details.
architectura
l drawings.
 LITERATURE SURVEY
S.NO Author Title Dataset Classification Merits Demerits Accuracy
Algorithm

9 Wang, Deep Floor RNN Tackles Limited by 85%

Learning plans, depth insufficient
Rocco
Approaches CAD data ambiguity ground truth
Furferi for 3D Model and training
Generation improves datasets
from 2D 3D model
Images geometry
 LITERATURE SURVEY
S.NO Author Title Dataset Classification Merits Demerits Accuracy
Algorithm

10 Jacob From Flat to Engineeri Rule-based and Evaluates Comparativel 84%

Sam, Spatial: ng and hybrid methods multiple y lower
Karan Comparison CAD methods scalability for
patel, of 4 Methods drawings for 3D larger
Mike Constructing model datasets
Saad 3D Models creation
from 2D
Plans
 REQUIRMENTS

Software Requiments:-
Hardware Requirments:-
• PYTHON IDLE
• Laptop with 8GB of R A M
• VS Code
• Processor (Intel i7)
• Flask, opencv-python
• G P U (1060)
• numpy, trimesh

• Pythreejs, Pillow
GOALS &
CHALLENGES
Goal: Generate accurate 3D models from 2D blueprints.

Challenges:

• Lack of depth information and hidden elements, making it difficult to infer the complete 3D
structure from a flat representation.

• Variations in blueprint styles, symbols, and notations across different industries and designers
can hinder a one-size-fits-all approach.

• Missing details or incomplete blueprints pose a significant obstacle, requiring intelligent

reconstruction techniques.

• Complex structures with intricate geometries increase computational complexity and demand
high processing power.

• Ensuring real-time processing and scalability for large-scale applications while maintaining high
accuracy remains a critical challenge.
DESIGN
DESIGN
 PROPOSED
METHODOLOGY
• Preprocessing: Blueprint images are enhanced for clarity.
• Image Segmentation: Structural elements are isolated.
• Feature Extraction: CNNs and GANs extract features from 2D blueprints.
• 3D Conversion: Features are converted into 3D representations.
• Model Training: Trained on datasets of 2D blueprints and corresponding 3D
models.
• Optimization: Techniques refine the 3D model's accuracy.
• Post-Processing: Scaling and rendering prepare the model for visualization.
• Iterative Refinement: Experimental evaluations enhance accuracy and
efficiency.
INPUT
Input of the 2d blueprints:
OUTPUT
DATASET
 DATASET
Floor Plan dataset:
 https://www.kaggle.com/datasets/asutoshprad/floor-plan-dataset

This dataset is a customized version of the raw CVC FPP dataset. The original dataset contained SVG
masked images, which have been converted to PNG format for easier handling and compatibility with
various image processing and machine learning frameworks.

 DATASET FILES
• Original Images: High-quality images from the CVC FPP dataset.
• PNG Masks: Converted from SVG format to PNG,
these masks are used for segmentation tasks.
CONCLUSION
 CONCLUSION
• Robust Methodology: Converts 2D blueprints into 3D models using neural
networks.
• Deep Learning Techniques: Extracts structural features and reconstructs
accurate 3D representations.
• Effective Automation: Validated experimentally, improving accuracy and
efficiency.
• Industry Impact: Aids researchers, designers, and professionals across
various domains.
REFERENCES
 REFERENCES
REFERENCES

• [1] Jia Zheng: cornell University ,United States .Yifan Zhu: cornell University , United
States. Submission by November 2022, https://arxiv.org/abs/2209.02692

• [2] Sofia Feist: University College Londan.Luis Jacques de Souza: University of

Porto.Luis Sanhudo: BUILT CoLAB.Joao pocas Martins: University of
Porto.Submitted by May 2024 ,
https://www.researchgate.net/publication/380440391_Automatic_Reconstruction_
of_3D_Models_from_2D_Drawings_A_State-of-the-Art_Review

• [3] Sungsoo Park : MDPI ,Basel,Switzerland. Hyeoncheol Kim : Department of

computer science and engineering,Korea University, seoul 02841,Korea.
Published in November 2021,
https://www.semanticscholar.org/paper/3DPlanNet%3A-Generating-3D-Models-fro
m-2D-Floor-Plan-Park-Kim/43c917eca8ae317e8da9c9200389e087e22673c8
• [4] Dr. Nasreen Fathima1, Mohammed Shezan2,Niha Rehaman3, Preksha Jain M4,
Shishira N51Associate Professor & Head, Department of Computer Science and
Design ATME College of Engineering, Mysore, India2,3,4,5 Department of Computer
Science and Design Student of ATME College of Engineering, Mysore,India| Volume
11 Issue 8 | ISSN: 2349-6002,
https://ijirt.org/publishedpaper/IJIRT171672_PAPER.pdf

• [5] Robin Courant : Computer vision and pattern Recognition in Cornell University,United
States.Xi Wang : Cornell University,United States. Marc Christie : computer science in
Cornell University, United States.Published in September 2023.
https://arxiv.org/abs/2309.03933

• [6] J.Ryan Shue,Ryan po,Zachary Ankner,Jiajun Wu,Gorden Wetzstein : Computer vision

and pattern Recognition in Cornell University, United States.Eric Ryan Chan : Computer
vision and pattern Recognition in Cornell University, United States.
https://arxiv.org/abs/2211.16677
• [7] Shoaib Ahmed Siddiqui: University of Cambridge. David Krueger: University of Cambridge.
Thomas Breuel : NVIDIA Research. Submitted by April 2023.https://arxiv.org/abs/2304.09358

• [8] Hamed Amiri, Hannah Vogel and Oliver Plumper:Department of Earth Sciences, Faculty of
Geosciences, Utrecht University, Utrecht, The Netherlands. RESEARCH
ARTICLE10.1029/2024JH000178
• https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2024JH000178

• [9] Rocco Furferi : Department of Industrial Engineering, University of Florence, 50139 Firenze,
Italy.Published: 28 December 2024. https://www.mdpi.com/2571-9408/8/1/12

• [10] Jacob Sam1, Karan Patel2, Mike Saad : Applied Research, scale XL, Image to 3d, Mesh
Reconstruction, Generative Design, computer science , graphics in Cornell University, United
States. Published in 29th July 2024. https://www.mdpi.com/2571-9408/8/1/12
THANK
YOU