I tried to solve the problem of computer vision about the detection of cars. Decided as a training (for myself), so my solution may not be the most optimal and best, but you can use it as a baseline in some competition or a simple CV pipeline.

I got data from this competition.

Fine tuning a fasterrcnn_resnet50_fpn about 5 epoch and I got not bad resault:

You can try it yourself with the show_img_car_detection.ipynb file. Мy weights.

If you want to see the process of creating and training, you can follow the link to my kaggle notebook