Open boheumd opened 4 years ago
I trained the TSM-resnet50 with 8 segments on SomethingV1 datasets with optical flow inputs only. The single crop accuracy for the flow model is around 37% and the two-stream result is around 48%. Is this performance reasonable or not?
please How you run the code on somethingv1 I could not run it correctly can share the code, please
I trained the TSM-resnet50 with 8 segments on SomethingV1 datasets with optical flow inputs only. The single crop accuracy for the flow model is around 37% and the two-stream result is around 48%. Is this performance reasonable or not?