search

yrcong / STTran

Spatial-Temporal Transformer for Dynamic Scene Graph Generation, ICCV2021
MIT License
181 stars 34 forks source link

the Action Genome Dump frames of this paper #12

Closed Yassin-fan closed 2 years ago

Yassin-fan commented 2 years ago

In your readme, it is said to follow JingweiJ/ActionGenome to get the dataset. Could you please supply the version of your ffmpeg? For I followed JingweiJ/ActionGenome and get the empty folders of all the videos. By the way, I'd like to know whether you use 'python tools/dump_frames.py --all_frames' or just 'python tools/dump_frames.py' to get the dataset? Thanks for your work!

entalent commented 2 years ago

If you get empty folders, this is probably because there are some errors when executing the ffmpeg command. In line 36 of dump_frames.py, you can change -loglevel panic to -loglevel verbose to see the output of ffmpeg and know what happened when extracting the frames. You can also check whether the command in this line is correctly formatted.

Yassin-fan commented 2 years ago

If you get empty folders, this is probably because there are some errors when executing the ffmpeg command. In line 36 of dump_frames.py, you can change -loglevel panic to -loglevel verbose to see the output of ffmpeg and know what happened when extracting the frames. You can also check whether the command in this line is correctly formatted.

Thanks for your help！ I sloved this proplem. By the way, I ran the train.py code and it said that "fail on background " after loading word vectors from glove.6B.200d.pt. In my opinion, it is because that the word " background " can not be found in glove, but how to slove this problem? Or just ignore it and wait for the result?

yrcong commented 2 years ago

If you get empty folders, this is probably because there are some errors when executing the ffmpeg command. In line 36 of dump_frames.py, you can change -loglevel panic to -loglevel verbose to see the output of ffmpeg and know what happened when extracting the frames. You can also check whether the command in this line is correctly formatted.

Thanks for your help！ I sloved this proplem. By the way, I ran the train.py code and it said that "fail on background " after loading word vectors from glove.6B.200d.pt. In my opinion, it is because that the word " background " can not be found in glove, but how to slove this problem? Or just ignore it and wait for the result?

You could easily ignore this:) The "background" is not included in the GloVe model, but we also didn't use it.