train ResNet-18 on COCO, got error

[northeastsquare]

Hi, I train ResNet-18 on COCO according to README.md: train_nGPU=1 test_nGPU=1 model=resnet resnet_path=./data/models/resnet/resnet-18.t7 ./scripts/train_coco.sh By the way, I installed torch7 luajit, But train out of memory, so I cd ~/torch; TORCH_LUA_VERSION=LUA51 ./install.sh

I don't know this matter or not.

I got the following error:

Loading proposals at { 1 : "/home/sam/src/multipathnet/data/proposals/coco/sharpmask/train.t7" 2 : "/home/sam/src/multipathnet/data/proposals/coco/sharpmask/val.t7" } Done loading proposals

proposal images 123287

dataset images 118287

images 123287

nImages 118287
Loading proposals at { 1 : "/home/sam/src/multipathnet/data/proposals/coco/sharpmask/train.t7" 2 : "/home/sam/src/multipathnet/data/proposals/coco/sharpmask/val.t7" } Done loading proposals

proposal images 123287

dataset images 118287

images 123287

nImages 118287
/home/sam/torch/install/bin/lua: /home/sam/torch/install/share/lua/5.1/trepl/init.lua:384: ./loaders/loader.lua:39: expected cdata for arg #1 stack traceback: C: in function 'error' /home/sam/torch/install/share/lua/5.1/trepl/init.lua:384: in function 'require' train.lua:121: in main chunk C: in function 'dofile' .../torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk

the whole output is : d.txt

[northeastsquare]

I ran ./clean.sh, in ~/torch, then TORCH_LUA_VERSION=LUA51 ./install.sh and install the other package, but result is the same.

[szagoruyko]

@northeastsquare I pushed a patch for coco in https://github.com/facebookresearch/multipathnet/commit/e6b9e0dc68db5af4662be5f6c272c1db82ab514d, can you pull and try again?

[northeastsquare]

Yes, after pull, and use pure lua51 not luajit, and comment many 'tds', it start to train.Thank you.