AKASH2907 / pi-consistency-activity-detection

End-to-End Semi-Supervised Learning for Video Action Detection [CVPR 2022]
MIT License
34 stars 4 forks source link

Load annotation in file 'ucf_dataloader_eval.py' #10

Open DuongVanKhoa0811 opened 3 months ago

DuongVanKhoa0811 commented 3 months ago

In the 'ucf_dataloader_eval.py' file, specifically from lines 134 to 140, could you explain why only one annotation per video is loaded? I am concerned that this approach might lead to a degradation in performance during model evaluation. The model is trained to detect all actors (annotations) in the video, but if only one annotation is considered during evaluation, it might decrease the Intersection over Union (IoU) and result in lower performance metrics.