Closed JayYangSS closed 2 years ago
@JayYangSS I tried it but the results were not better....I guess it needs more work to attain further improvement.
I try the SSL training, it seems that the unlabel data does not contribute the performance gain. I think pseudo filter need to be designed for a further improvement
I found in
trainer_sessd.py
, line334-339, you comment the unlabeled data, Have you try to add the unlabeled data to your training process to finish the semi-supervised training? How's the effect of SSL?