not enough RAM

[atonn]

Hello, first of all thank you for uploading your training code!

Some issues I encountered:

1. train.lua was missing lib/tv.lua and lib/tv2.lua. I used the tv.lua from your cnn-mrf repository and commented out the line for tv2.

2. new users, don't forget to set flag_MDAN = true, flag_AG =true, flag_MGAN = true in train.lua if you actually want it to do the whole training process. Only the first step, flag_MDAN, is set true by default.

3. I had to set opt.batchSize in MGAN_wrapper.lua from 64 to 32 since my 4 GB GPU ran out of memory at this step.

4. In order to successfully run release_MGAN.lua, I had to manually move the file netS_1.t7 from Dataset/VG_Alpilles_ImageNet100/MDAN to Dataset/VG_Alpilles_ImageNet100/MGAN.

5. Finally, when running demo_MGAN.lua, I am stuck with:

`/home/atonn/torch/install/bin/luajit: /home/atonn/torch/install/share/lua/5.1/nn/Container.lua:67:
In 23 module of nn.Sequential:
/home/atonn/torch/install/share/lua/5.1/nn/THNN.lua:109: bad argument #2 to 'v' (input channels and nInputPlane dont match at /tmp/luarocks_cunn-scm-1-2642/cunn/lib/THCUNN/SpatialConvolutionMM.cu:29)
stack traceback:
[C]: in function 'v'
/home/atonn/torch/install/share/lua/5.1/nn/THNN.lua:109: in function 'SpatialConvolutionMM_updateOutput'
...id/torch/install/share/lua/5.1/nn/SpatialConvolution.lua:111: in function <...id/torch/install/share/lua/5.1/nn/SpatialConvolution.lua:107>
[C]: in function 'xpcall'
/home/atonn/torch/install/share/lua/5.1/nn/Container.lua:63: in function 'rethrowErrors'
/home/atonn/torch/install/share/lua/5.1/nn/Sequential.lua:44: in function 'forward'
demo_MGAN.lua:56: in main chunk
[C]: in function 'dofile'
...poid/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk
[C]: at 0x00406670

WARNING: If you see a stack trace below, it doesn't point to the place where this error occured. Please use only the one above.
stack traceback:
[C]: in function 'error'
/home/atonn/torch/install/share/lua/5.1/nn/Container.lua:67: in function 'rethrowErrors'
/home/atonn/torch/install/share/lua/5.1/nn/Sequential.lua:44: in function 'forward'
demo_MGAN.lua:56: in main chunk
[C]: in function 'dofile'
...poid/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk
[C]: at 0x00406670`

I suppose that's what I get for messing with the batch size? When substituting my MGAN file with the provided pretrained file it runs just fine. Any help would be appreciated. Cheers :D

[chuanli11]

Hi, atonn

Thanks a lot for the feedbacks. I uploaded the missing tv files and it should solve all your problems.

[atonn]

Thank you! All of my problems were indeed solved. The results don't look as good as those from your provided models, but I guess that is due to the batchSize of 32?

[hankobe]

HI, @atonn
How about now? Is the results you obtained good or not?