Sizes do not match error

[syed-ahmed]

Hi I tried to reproduce the training from this research paper. I was able to successfully train the encoder. However, I get the following error when training the decoder. Any help is appreciated.

Command used to train the encoder

th run.lua --dataset cv --datapath /home/ubuntu/SegNet-Tutorial/CamVid/ --model models/encoder.lua --save /home/ubuntu/ENet-training/train/trained_model/ --imHeight 360 --imWidth 480 --labelHeight 45 --labelWidth 60 --cachepath /home/ubuntu/ENet-training/train/dataset_cache/

Command used to start training the decoder and its resulting error:

th run.lua --dataset cv --datapath /home/ubuntu/SegNet-Tutorial/CamVid/ --model models/decoder.lua --save /home/ubuntu/ENet-training/train/trained_decoder/ --imHeight 360 --imWidth 480 --labelHeight 360 --labelWidth 480 --cachepath /home/ubuntu/ENet-training/train/dataset_cache/ --CNNEncoder /home/ubuntu/ENet-training/train/trained_model/model-299.net

Error:

==> Training: epoch # 1 [batchSize = 2] /home/ubuntu/torch/install/bin/luajit: bad argument #2 to '?' (sizes do not match at /home/ubuntu/torch/extra/cutorch/lib/THC/generic/THCTensorCopy.c:64) stack traceback: [C]: at 0x7ff8eee7d610 [C]: in function '__newindex' ./train.lua:97: in function 'train' run.lua:77: in main chunk [C]: in function 'dofile' ...untu/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk [C]: at 0x00406670

Another command used to start training the decoder and its resulting error:

th run.lua --dataset cv --datapath /home/ubuntu/SegNet-Tutorial/CamVid/ --model models/decoder.lua --save /home/ubuntu/ENet-training/train/trained_decoder/ --imHeight 45 --imWidth 60 --labelHeight 360 --labelWidth 480 --cachepath /home/ubuntu/ENet-training/train/dataset_cache/ --CNNEncoder /home/ubuntu/ENet-training/train/trained_model/model-299.net

Error:

==> Training: epoch # 1 [batchSize = 2] /home/ubuntu/torch/install/bin/luajit: /home/ubuntu/torch/install/share/lua/5.1/nn/Container.lua:67: Step: 0ms In 2 module of nn.Sequential: /home/ubuntu/torch/install/share/lua/5.1/nn/JoinTable.lua:39: bad argument #1 to 'copy' (sizes do not match at /home/ubuntu/torch/extra/cutorch/lib/THC/generic/THCTensorCopy.cu:10) stack traceback: [C]: in function 'copy' /home/ubuntu/torch/install/share/lua/5.1/nn/JoinTable.lua:39: in function </home/ubuntu/torch/install/share/lua/5.1/nn/JoinTable.lua:21> [C]: in function 'xpcall' /home/ubuntu/torch/install/share/lua/5.1/nn/Container.lua:63: in function 'rethrowErrors' /home/ubuntu/torch/install/share/lua/5.1/nn/Sequential.lua:44: in function 'forward' ./train.lua:108: in function 'opfunc' /home/ubuntu/torch/install/share/lua/5.1/optim/adam.lua:33: in function 'adam' ./train.lua:123: in function 'train' run.lua:77: in main chunk [C]: in function 'dofile' ...untu/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk [C]: at 0x00406670

[codeAC29]

You are using same cachepath for encoder and decoder in the first approach. Try using a different location of datacache. Currently its using the dataset which it saved while training encoder.

Image and label resolution needs to be the same for decoder, so second approach is incorrect.

[syed-ahmed]

Thanks! That worked :)