LOSS stuck arount 0.69

[ashishjamarkattel]

I got an strange issue while training color_sync net. I have read your other issue for filtering the dataset. to [-1, 1 ] i really dont understand this. How do we do that.

1. I have cut the video into less than 5sec.
2. resolution of video is 760px to 1080px.
3. Lr is 1e-4 also tried 1e-5
4. batch size is 16.

[NochnoyRitzar]

Hi! [-1, 1] refers to are offset values between audio and video that you get after using code from https://github.com/joonson/syncnet_python on your videos.

I recommend running run_pipeline.py first, as it preprocesses video to make it suitable for syncnet, and then running run_syncnet.py from the same repo

[ashishjamarkattel]

Thanks for the help. Is running those 2 command is enough ?

[NochnoyRitzar]

Yes. There is another one which is run_visualize.py, but it's just for visualization purpose

[397688551]

I have encountered the same problem, which is also not falling at 0.69. May I ask if you have solved it？

[ashishjamarkattel]

I have encountered the same problem, which is also not falling at 0.69. May I ask if you have solved it？

I am working on it. But i am not able to solve it still. Somewhere in case of mine syncnet_python not working for my avspeech dataset. Please let me know if you have solve it.

[drakitLiu]

I have encountered the same problem, which is also not falling at 0.69. May I ask if you have solved it？

I am working on it. But i am not able to solve it still. Somewhere in case of mine syncnet_python not working for my avspeech dataset. Please let me know if you have solve it. I have encountered the same problem, which is also not falling at 0.69. oh my ! May I ask if you have solved it？

[drakitLiu]

I have encountered the same problem, which is also not falling at 0.69. May I ask if you have solved it？

I am working on it. But i am not able to solve it still. Somewhere in case of mine syncnet_python not working for my avspeech dataset. Please let me know if you have solve it.

Hi ! did you have solved it？