Gorilla-Lab-SCUT/frustum-convnet

Difference between car and car_refine?

sarimmehdi opened this issue · 1 comments

Hello. I ran your code by typing bash scripts/eval_pretrained_models.sh and I notice two folders created inside the pretrained_models folder. In the car folder, I have detection files for all image frames but in car_refine I have fewer detection files. Is there any reason for this? Are the detections in car_refine more accurate?

Yes, car_refine will have more accurate results. The differences of them are different proposal space and some config settings. car is based on initial 2D detection proposals but car_refine is based on the first stage predicted results .