Support for RGBPose_Conv3D on UCF101 and HMDB51

[Narrowunknown]

Thank you for your contributions! I noticed that your paper Revisiting Skeleton-based Action Recognition has witnessed a better overall performance on UCF101 and HMDB51 after fusing with predictions based on other modalities(like RGB). But there are not yet weights of RGBPose_Conv3D on UCF101 and HMDB51, would you please release these pretrained weights so that we can make some validations? We will sincerely appreciate it!