search

chenxuluo / GST-video

ICCV 19 Grouped Spatial-Temporal Aggretation for Efficient Action Recognition
MIT License
43 stars 11 forks source link

About S3D/C3D #2

Open FloydEdwin opened 4 years ago

FloydEdwin commented 4 years ago

Hello! Thank you for sharing your code.Great work you are.I have two questions to ask you.

  1. I read the code in models.py ######### elif model == 'C3D': import C3D self.base_model = getattr(C3D, backbone)() elif model == 'S3D': import S3D self.base_model = getattr(S3D, backbone)() ########## I find the base_model of S3D/C3D are not exist.I would appreciate it if you willing to share C3D or S3D.
  2. I find your work only pretrained on Imagenet.I have seen many work using Kinetics pretrained （such as I3D）.Have you tried and pretrained on Kinetics? If not , why not? Thank you !
slighting666 commented 3 years ago

@chenxuluo Hello, did you succeed in embedding these two branches in resnet? I tried this way according to the author, but I always get this error‘RuntimeError: The size of tensor a (28) must match the size of tensor b (14) at non-singleton dimension 4’. Can you help me, I need your help. qq: 858856173. Thank you!

slighting666 commented 3 years ago

@K-23333 hi，do you know the version GST of Resnet-18? thanks!