search

open-mmlab / mmaction2

OpenMMLab's Next Generation Video Understanding Toolbox and Benchmark
https://mmaction2.readthedocs.io
Apache License 2.0
4.29k stars 1.25k forks source link

Structure of a model in the input layers. #2111

Open antonio2600 opened 1 year ago

antonio2600 commented 1 year ago

Hello. How can I see the structure that a model follows in the input layers? I want to understand well how a model trains and also test it with my own dataset, so I have a doubt about the duration of the videos that the model accepts. For example, I want to know how long are the videos that accepts tsn_r50_video_1x1x8_100e_kinetics400_rgb.py.

hukkai commented 1 year ago

tsn_r50_video_1x1x8_100e_kinetics400_rgb.py accepts videos of any length, best if the video has more than 8 frames. Our codes have handled the case where videos are short.

antonio2600 commented 1 year ago

Ok. My dataset has got videos of diferent length, 1min 30s, 1min, 30s, 10s. Does this afect network training?

hukkai commented 1 year ago

From my idea, not really.