Closed lusinlu closed 2 years ago
The solution is the following - there is a bug in the trajectories dataset and in csv files rows with indexes 1 and 2 are duplicates, which is not correct!!! For anyone who would use the training data with video frames - you should remove the duplicate to match the video frames
Hi! Could you please tell, me what library you used to extract frames from original ShanghaiTech videos? Currently, Im using OpenCV (cv2.VideoCapture) but its returning fewer frames compared with the frame ids that you have in csv files.