Road damage detection

[Adhavan1801]

Pull Request for DL-Simplified 💡

Issue Title : Add Multi-Modal Deep Learning Framework for Road Damage Detection

• Info about the related issue (Aim of the project) : Build, train, and evaluate multiple deep learning architectures (EfficientNet-B0, ResNet50, YOLOv8n, Vision Transformer ViT-B/16) on the RDD-2022 dataset to detect and classify 4 types of road damage (Longitudinal Crack, Transverse Crack, Alligator Crack, Pothole) with model comparison and Grad-CAM visualizations.
• Name: Adhavan
• GitHub ID: Adhavan1801
• Email ID: adhavan1801@gmail.com
• Identify yourself: GSSoC 2026 Participant

Closes: #1074

Describe the add-ons or changes you've made 📃

Added a complete Multi-Modal Road Damage Detection project under Multi-Modal Road Damage Detection/ with the following:

• Jupyter Notebook (Model/Road_Damage_Detection.ipynb) implementing 4 deep learning models:

  • EfficientNet-B0 — 4.01M params, 71.31% test accuracy
  • ResNet50 — 25M params, 75.86% test accuracy (best classifier)
  • YOLOv8n — object detection with bounding box localization (7.52ms/img fastest latency)
  • Vision Transformer (ViT-B/16) — 86M params, 73.42% test accuracy

• Dataset: RDD-2022 (Road Damage Dataset) with 26,869 train / 5,758 val / 5,758 test images across 4 damage classes (D00, D10, D20, D40) from multiple countries

• Training features: Early stopping (patience=5), Cosine Annealing LR scheduler, GPU-accelerated training (RTX 5060, 8.5GB VRAM)

• Grad-CAM visualizations for all 3 classification models to explain predictions

• Model results CSV (Model/model_results_summary.csv) with accuracy, precision, recall, F1-score, and inference latency for all models

• Comprehensive README.md with dataset details, model architectures, training pipeline, EDA plots, results table, and setup instructions

Type of change ☑️

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Code style update (formatting, local variables)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• This change requires a documentation update

How Has This Been Tested? ⚙️

• All 4 models were fully trained and evaluated on the RDD-2022 test set (5,758 images)
• Classification metrics (accuracy, precision, recall, F1-score) computed using scikit-learn
• Inference latency measured per image on NVIDIA RTX 5060 GPU
• Grad-CAM visualizations generated and verified for all 3 CNN/ViT models
• Notebook executed end-to-end without errors in a local virtual environment with CUDA 12.8

Test Results:

Model	Accuracy	F1-Score	Latency
EfficientNet-B0	71.31%	0.7131	8.48 ms
ResNet50	75.86%	0.7573	8.64 ms
YOLOv8n	56.59%	0.5707	7.52 ms
ViT-B/16	73.42%	0.7275	12.95 ms

Checklist: ☑️

• My code follows the guidelines of this project.
• I have performed a self-review of my own code.
• I have commented my code, particularly wherever it was hard to understand.
• I have made corresponding changes to the documentation.
• My changes generate no new warnings.
• I have added things that prove my fix is effective or that my feature works.
• Any dependent changes have been merged and published in downstream modules.

[github-actions]

Our team will soon review your PR. Thanks @Adhavan1801 :)