Computer Vision Challenge

This project fine tunes a pretrained ResNet-152 model for Car Model Recognition.

Code snippets for Car Detection with YOLOv3 has been adopted from this project (https://github.com/eriklindernoren/PyTorch-YOLOv3.git)

Download data and model weights

• Download necessary files

bash ./download.sh

• Download best.tar, the model trained for the project from Google Drive and put under ./model/ directory

https://drive.google.com/open?id=1B4NZOXJS2lMjYjloPmY-JGIrTBL5p6Dx

Prepare training data

• Preprocess the data
  • crop the car images defined by the bouding boxes
  • split them into training/validation datasets with the ratio of 8:2.

python ./load_data.py

To Train

• Train with ResNet-152 as base model.

python ./train.py

To Test

• Test accuracy on validation dataset

python ./train.py --validation

To Predict

• Run prediction on a single image (cropped car image)

python ./predict.py --mode predict --image_path <image_path>

• Run detection and prediction on a single image (not cropped) with visualization

python ./predict.py --mode predict_and_detect --image_path <image_path>

TODO

• Replace ResNet-152 with EfficientNet (https://arxiv.org/abs/1905.11946) (implemented with PyTorch in the project, not trained yet)