search

szagoruyko / imagine-nn

IMAGINE torch neural network routines
Other
109 stars 35 forks source link

encountering error when converting resnet-50 torch model by facebook #45

Open michaelhuang74 opened 7 years ago

michaelhuang74 commented 7 years ago

When I used the script on https://gist.github.com/szagoruyko/8828e09cc4687afd324d, which used the utils.lua, to convert the facebook resnet-50 model file (https://github.com/facebook/fb.resnet.torch), I had the following error:

Couldnt fold these: { 1 : { gradBias : CudaTensor - size: 256 bias : CudaTensor - size: 256 output : CudaTensor - size: 10x256x56x56 gradInput : CudaTensor - empty save_mean : CudaTensor - size: 256 running_var : CudaTensor - size: 256 momentum : 0.1 eps : 1e-05 weight : CudaTensor - size: 256 _type : "torch.CudaTensor" affine : true running_mean : CudaTensor - size: 256 gradWeight : CudaTensor - size: 256 save_std : CudaTensor - size: 256 train : false } 2 : { gradBias : CudaTensor - size: 256 bias : CudaTensor - size: 256 output : CudaTensor - size: 10x256x56x56 gradInput : CudaTensor - empty save_mean : CudaTensor - size: 256 running_var : CudaTensor - size: 256 momentum : 0.1 eps : 1e-05 weight : CudaTensor - size: 256 _type : "torch.CudaTensor" affine : true running_mean : CudaTensor - size: 256 gradWeight : CudaTensor - size: 256 save_std : CudaTensor - size: 256 train : false } 3 : { gradBias : CudaTensor - size: 256 bias : CudaTensor - size: 256 output : CudaTensor - size: 10x256x56x56 gradInput : CudaTensor - empty save_mean : CudaTensor - size: 256 running_var : CudaTensor - size: 256 momentum : 0.1 eps : 1e-05 weight : CudaTensor - size: 256 _type : "torch.CudaTensor" affine : true running_mean : CudaTensor - size: 256 gradWeight : CudaTensor - size: 256 save_std : CudaTensor - size: 256 train : false } 4 : { gradBias : CudaTensor - size: 512 bias : CudaTensor - size: 512 output : CudaTensor - size: 10x512x28x28 gradInput : CudaTensor - empty save_mean : CudaTensor - size: 512 running_var : CudaTensor - size: 512 momentum : 0.1 eps : 1e-05 weight : CudaTensor - size: 512 _type : "torch.CudaTensor" affine : true running_mean : CudaTensor - size: 512 gradWeight : CudaTensor - size: 512 save_std : CudaTensor - size: 512 train : false } 5 : { gradBias : CudaTensor - size: 512 bias : CudaTensor - size: 512 output : CudaTensor - size: 10x512x28x28 gradInput : CudaTensor - empty save_mean : CudaTensor - size: 512 running_var : CudaTensor - size: 512 momentum : 0.1 eps : 1e-05 weight : CudaTensor - size: 512 _type : "torch.CudaTensor" affine : true running_mean : CudaTensor - size: 512 gradWeight : CudaTensor - size: 512 save_std : CudaTensor - size: 512 train : false } 6 : { gradBias : CudaTensor - size: 512 bias : CudaTensor - size: 512 output : CudaTensor - size: 10x512x28x28 gradInput : CudaTensor - empty save_mean : CudaTensor - size: 512 running_var : CudaTensor - size: 512 momentum : 0.1 eps : 1e-05 weight : CudaTensor - size: 512 _type : "torch.CudaTensor" affine : true running_mean : CudaTensor - size: 512 gradWeight : CudaTensor - size: 512 save_std : CudaTensor - size: 512 train : false } 7 : { gradBias : CudaTensor - size: 512 bias : CudaTensor - size: 512 output : CudaTensor - size: 10x512x28x28 gradInput : CudaTensor - empty save_mean : CudaTensor - size: 512 running_var : CudaTensor - size: 512 momentum : 0.1 eps : 1e-05 weight : CudaTensor - size: 512 _type : "torch.CudaTensor" affine : true running_mean : CudaTensor - size: 512 gradWeight : CudaTensor - size: 512 save_std : CudaTensor - size: 512 train : false } 8 : { gradBias : CudaTensor - size: 1024 bias : CudaTensor - size: 1024 output : CudaTensor - size: 10x1024x14x14 gradInput : CudaTensor - empty save_mean : CudaTensor - size: 1024 running_var : CudaTensor - size: 1024 momentum : 0.1 eps : 1e-05 weight : CudaTensor - size: 1024 _type : "torch.CudaTensor" affine : true running_mean : CudaTensor - size: 1024 gradWeight : CudaTensor - size: 1024 save_std : CudaTensor - size: 1024 train : false } 9 : { gradBias : CudaTensor - size: 1024 bias : CudaTensor - size: 1024 output : CudaTensor - size: 10x1024x14x14 gradInput : CudaTensor - empty save_mean : CudaTensor - size: 1024 running_var : CudaTensor - size: 1024 momentum : 0.1 eps : 1e-05 weight : CudaTensor - size: 1024 _type : "torch.CudaTensor" affine : true running_mean : CudaTensor - size: 1024 gradWeight : CudaTensor - size: 1024 save_std : CudaTensor - size: 1024 train : false } 10 : { gradBias : CudaTensor - size: 1024 bias : CudaTensor - size: 1024 output : CudaTensor - size: 10x1024x14x14 gradInput : CudaTensor - empty save_mean : CudaTensor - size: 1024 running_var : CudaTensor - size: 1024 momentum : 0.1 eps : 1e-05 weight : CudaTensor - size: 1024 _type : "torch.CudaTensor" affine : true running_mean : CudaTensor - size: 1024 gradWeight : CudaTensor - size: 1024 save_std : CudaTensor - size: 1024 train : false } 11 : { gradBias : CudaTensor - size: 1024 bias : CudaTensor - size: 1024 output : CudaTensor - size: 10x1024x14x14 gradInput : CudaTensor - empty save_mean : CudaTensor - size: 1024 running_var : CudaTensor - size: 1024 momentum : 0.1 eps : 1e-05 weight : CudaTensor - size: 1024 _type : "torch.CudaTensor" affine : true running_mean : CudaTensor - size: 1024 gradWeight : CudaTensor - size: 1024 save_std : CudaTensor - size: 1024 train : false } 12 : { gradBias : CudaTensor - size: 1024 bias : CudaTensor - size: 1024 output : CudaTensor - size: 10x1024x14x14 gradInput : CudaTensor - empty save_mean : CudaTensor - size: 1024 running_var : CudaTensor - size: 1024 momentum : 0.1 eps : 1e-05 weight : CudaTensor - size: 1024 _type : "torch.CudaTensor" affine : true running_mean : CudaTensor - size: 1024 gradWeight : CudaTensor - size: 1024 save_std : CudaTensor - size: 1024 train : false } 13 : { gradBias : CudaTensor - size: 1024 bias : CudaTensor - size: 1024 output : CudaTensor - size: 10x1024x14x14 gradInput : CudaTensor - empty save_mean : CudaTensor - size: 1024 running_var : CudaTensor - size: 1024 momentum : 0.1 eps : 1e-05 weight : CudaTensor - size: 1024 _type : "torch.CudaTensor" affine : true running_mean : CudaTensor - size: 1024 gradWeight : CudaTensor - size: 1024 save_std : CudaTensor - size: 1024 train : false } 14 : { gradBias : CudaTensor - size: 2048 bias : CudaTensor - size: 2048 output : CudaTensor - size: 10x2048x7x7 gradInput : CudaTensor - empty save_mean : CudaTensor - size: 2048 running_var : CudaTensor - size: 2048 momentum : 0.1 eps : 1e-05 weight : CudaTensor - size: 2048 _type : "torch.CudaTensor" affine : true running_mean : CudaTensor - size: 2048 gradWeight : CudaTensor - size: 2048 save_std : CudaTensor - size: 2048 train : false } 15 : { gradBias : CudaTensor - size: 2048 bias : CudaTensor - size: 2048 output : CudaTensor - size: 10x2048x7x7 gradInput : CudaTensor - empty save_mean : CudaTensor - size: 2048 running_var : CudaTensor - size: 2048 momentum : 0.1 eps : 1e-05 weight : CudaTensor - size: 2048 _type : "torch.CudaTensor" affine : true running_mean : CudaTensor - size: 2048 gradWeight : CudaTensor - size: 2048 save_std : CudaTensor - size: 2048 train : false } 16 : { gradBias : CudaTensor - size: 2048 bias : CudaTensor - size: 2048 output : CudaTensor - size: 10x2048x7x7 gradInput : CudaTensor - empty save_mean : CudaTensor - size: 2048 running_var : CudaTensor - size: 2048 momentum : 0.1 eps : 1e-05 weight : CudaTensor - size: 2048 _type : "torch.CudaTensor" affine : true running_mean : CudaTensor - size: 2048 gradWeight : CudaTensor - size: 2048 save_std : CudaTensor - size: 2048 train : false } }

Any idea how to resolve it?