few-shot-action-recognition
few-shot-action-recognition copied to clipboard
questions about split of TRX
I reimplementation TRX on hmdb51,however, the accuracy of the test set is always several percentage points higher than that of the test set(about 5-10 percentage points).And the average accuracy of the test and validation sets is almost equal to the value reported in the TRX.Are you the same?If not,could you tell me the possible reasons?Thank you.