gradOutput size mismatch
Viresh-R opened this issue · 0 comments
Viresh-R commented
Hey,
I am using the fast RCNN implementation on a different dataset. When using the model defined in frcnn_alexnet.lua and passing as input "input=torch.Tensor(1,3,224,224):cuda()", I get the following error during the backward pass: "THNN.lua:110: Need gradOutput of dimension 3 and gradOutput.size[0] == 96 but got gradOutput to be of shape: [1 x 96 x 56 x 56] at /tmp/luarocks_cunn-scm-1-5180/cunn/lib/THCUNN/generic/SpatialConvolutionMM.cu:52 ". The forward pass works without any errors.