Print out confusion matrix in training and testing

[nixczhou]

I have a question, how do I modify to code in order to print out the confusion matrix in training and testing?

[innerlee]

You can implement it as a kind of evaluation metric. Similar to how mean_class_accuracy is done https://github.com/open-mmlab/mmaction2/blob/master/mmaction/datasets/base.py#L202

[irvingzhang0512]

If you try to show other metrics during training(along with top1_acc, top5_acc, loss, etc), you should modify head. For recognizers, you should modify BaseHead, add keys for dict losses