Could you please provide an example to generate the pkl files? Now it is very confusing to participants. Also, if possible, could you please give detailed step-by-step instructions for the data processing and generating the submission files. The github and the introduction in the webpage are completely messy. No clear logic is provided.
Posted by: wang3702 @ April 17, 2021, 3:33 a.m.Sincerely apologize for the inconvenience. We will provide a step-by-step instruction as soon as possible.
Stan
Posted by: leiwx52 @ April 18, 2021, 1 a.m.Dear participant,
We have updated our Github repo to provide instructions for preparing data and submission files: https://github.com/StanLei52/GEBD/blob/main/INSTRUCTIONS.md
Stan
Posted by: leiwx52 @ April 20, 2021, 12:42 p.m.How to prepare the test submission since we don't have the fps. Can we use the duration in the original test.json to calculate it from frames to the timestamp?
Posted by: wang3702 @ April 21, 2021, 11:03 p.m.You can get the fps for each video easily by using opencv:
################################
import cv2
cap = cv2.VideoCapture(args.video_path)
fps = cap.get(cv2.CAP_PROP_FPS)
################################
Actually you'd better check and use the fps of your downloaded videos over the train set, val set and test set in case there exists different fps between your videos and ours :)
As long as you download and trim the videos and use our annotations/boundary timestamps (in sec) correctly, it should be fine.