-
Notifications
You must be signed in to change notification settings - Fork 1.3k
Closed
Description
With the commit prior to the unsupervised additions, I am able to easily get 70+% top-1 accuracy using the Kinetics/SLOW_8x8_R50.yaml finetuning config. But with the new unsupervised code, I can barely get 40% top-1 accuracy on Kinetics-400 using the same config and training command.
There are a number of bugs I had to fix for the code to run, but I think there may be more. Could the authors consider carefully examining the new codebase for bugs that were introduced with recent code?
- PYAV issue (see Fix decoding issue with PYAV due to new support for multiple training… #541)
- LR scaling no longer works. With 99a655bd533d7fddd7f79509e3dfaae811767b5c, the call to assert_and_infer_cfg() was removed from
run_net.py. Thus LR scaling doesn't work as nowhere else in the code scales the LR by the number of shards.
I am also unable to reproduce the results of the MoCo unsupervised model trained with p=2 for 200 epochs, but there are larger issues if supervised R50 no longer works on Kinetics either.
Metadata
Metadata
Assignees
Labels
No labels