rpegoud/pytorch_traffic_sign_detection

Traffic sign image classification on the GTSRB dataset using PyTorch

GTRSB: Building a traffic sign recognition CNN using PyTorch

This project aims to train a PyTorch model to detect and identify traffic signs.

The images used come from the GTRSB (German Traffic Sign Recognition Benchmark) dataset, available on TorchVision.

The project was carried out in five steps:

• 1: Data exploration, creation of a simple baseline model (see: 1_baseline.ipynb)
• 2: Creation of a more complex CNN, improving on the baseline (see: 2_improving_CNN.ipynb)
• 3: Verifying the performances with cross validation (see: 3_cross_validation.ipynb)
• 4: Comparing our CNN with a pretrained EfficientNet (see: 4_transfer_learning.ipynb)
• 5: Implementing early stopping (see: 5_early_stopping.ipynb)

Results

Model	Accuracy (%)	Macro F1-score (%)	Number of training epochs	Training time per epoch (s)
Baseline	89	84	10	17
Best CNN	94	92	10	30
EfficientNet V2_small	99	98	10	120
EfficientNet V2_small early_stopping	97	95	2	120

General conclusions

In these notebooks, we obtained up to 99% accuracy and 98% macro F1-score on the task of traffic signs recognition.

Here are some ways to take this project further:

• The images are cropped and zoomed on specific traffic signs, a next step could be to collect images from road situations and detect traffic signs when there are some
• Use the coordinates features available in the dataset to train a segmentation model
• Apply segmentation to videos

Dataset Credits :

J. Stallkamp, M. Schlipsing, J. Salmen, and C. Igel. The German Traffic Sign Recognition Benchmark: A multi-class classification competition. In Proceedings of the IEEE International Joint Conference on Neural Networks, pages 1453–1460. 2011.

@inproceedings{Stallkamp-IJCNN-2011, author = {Johannes Stallkamp and Marc Schlipsing and Jan Salmen and Christian Igel}, booktitle = {IEEE International Joint Conference on Neural Networks}, title = {The {G}erman {T}raffic {S}ign {R}ecognition {B}enchmark: A multi-class classification competition}, year = {2011}, pages = {1453--1460} }