Pre-training model import errors and training issues

[xingjiy]

Why is it that after following the process of downloading the pre-trained model and then starting to train the CSL-Daily dataset, the message “!!!! pred_b: can not be converted, got head.projection.bias
pred_b: can not be converted, got head.projection.bias !!! pred_w: can not be converted, got head.projection.weight", and then the training stops automatically at the 85th example without any error report!

[junseok520]

We apologize for the delay in responding.

(1) The messages between 'Use Bidirectional Add Fusion' and 'Loading model finished' occur during the process of loading a pre-trained SlowFast model into a SlowFastSign model. These messages are caused by the difference in dimensions between the two models and the presence of modules, so there is nothing wrong with them and you can ignore them.

(2) In our training environment, the training does not stop at the 85th sample, please debug and let us know where you expect it to stop.

Thank you.

[zhang-prayer]

Why is it that after following the process of downloading the pre-trained model and then starting to train the CSL-Daily dataset, the message “!!!! pred_b: can not be converted, got head.projection.bias pred_b: can not be converted, got head.projection.bias !!! pred_w: can not be converted, got head.projection.weight", and then the training stops automatically at the 85th example without any error report!

Have you solved this problem yet?