search

facebookresearch / multipathnet

A Torch implementation of the object detection network from "A MultiPath Network for Object Detection" (https://arxiv.org/abs/1604.02135)
Other
1.34k stars 275 forks source link

Error parsing file names in loader.lua #38

Open outgrabe opened 7 years ago

outgrabe commented 7 years ago

Residual issue from #33

On Ubuntu 14.04, Torch with Lua 5.1.

train_nGPU=2 test_nGPU=1 ./scripts/train_multipathnet_coco.sh 
{
  phase2_learningRate : -1
  resume : ""
  weightDecay : 0
  learningRate : 0.001
  step : 2800
  bg_threshold_min : 0.1
  train_set : "trainval"
  train_min_gtroi_size : 0
  test_nsamples : 1000
  train_nGPU : 2
  learningRateDecay : 0
  test_num_per_image : 100
  test_nGPU : 1
  epoch : 1
  disable_memory_efficient_forward : false
  batchSize : 64
  extra_proposals_file : ""
  year : "2014"
  model : "multipathnet"
  dampening : 0
  nEpochs : 3200
  train_remove_dropouts : false
  manualSeed : 555
  imagenet_classes : ""
  retrain : "no"
  save_folder : "logs/coco_multipathnet_sharpmask_328225696"
  criterion : "ce"
  scale : 800
  bbox_regression : 1
  images_per_batch : 4
  nDonkeys : 6
  decay : 0.1
  dataset : "coco"
  best_proposals_number : 1000
  retrain_mean_std : ""
  epochSize : 100
  phase2_step : -1
  train_nsamples : -1
  fg_threshold : 0.5
  bg_threshold_max : 0.5
  phase2_decay : -1
  method : "sgd"
  test_set : "val"
  test_best_proposals_number : 400
  snapshot : 100
  sample_n_per_box : 0
  proposal_dir : "data/proposals/"
  integral : true
  proposals : "sharpmask"
  train_min_proposal_size : 0
  checkpoint : false
  sample_sigma : 1
  phase2_epoch : -1
  max_size : 1000
  momentum : 0.9
}
not found: THNN_CudaHalfELU_updateOutput...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type 
not found: THNN_CudaHalfELU_updateGradInput...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type  
not found: THNN_CudaHalfHardTanh_updateOutput...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type    
not found: THNN_CudaHalfHardTanh_updateGradInput...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type 
not found: THNN_CudaHalfLeakyReLU_updateOutput...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type   
not found: THNN_CudaHalfLeakyReLU_updateGradInput...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type    
not found: THNN_CudaHalfLookupTable_accGradParameters...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type    
not found: THNN_CudaHalfLookupTable_renorm...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type   
not found: THNN_CudaHalfMarginCriterion_updateOutput...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type 
not found: THNN_CudaHalfMarginCriterion_updateGradInput...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type  
not found: THNN_CudaHalfMultiMarginCriterion_updateOutput...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type    
not found: THNN_CudaHalfMultiMarginCriterion_updateGradInput...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type 
not found: THNN_CudaHalfPReLU_accGradParameters...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type  
not found: THNN_CudaHalfSpatialConvolutionLocal_accGradParameters...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type    
not found: THNN_CudaHalfSpatialConvolutionMM_accGradParameters...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type   
not found: THNN_CudaHalfSpatialCrossMapLRN_updateOutput...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type  
not found: THNN_CudaHalfSpatialCrossMapLRN_updateGradInput...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type   
not found: THNN_CudaHalfSpatialDilatedConvolution_accGradParameters...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type  
not found: THNN_CudaHalfSpatialFullConvolution_accGradParameters...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type 
not found: THNN_CudaHalfSoftPlus_updateOutput...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type    
not found: THNN_CudaHalfSoftPlus_updateGradInput...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type 
not found: THNN_CudaHalfSoftShrink_updateOutput...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type  
not found: THNN_CudaHalfSoftShrink_updateGradInput...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type   
not found: THNN_CudaHalfSqrt_updateOutput...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type    
not found: THNN_CudaHalfTemporalConvolution_accGradParameters...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type    
not found: THNN_CudaHalfThreshold_updateOutput...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type   
not found: THNN_CudaHalfThreshold_updateGradInput...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type    
not found: THNN_CudaHalfVolumetricConvolution_accGradParameters...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type  
not found: THNN_CudaHalfVolumetricDilatedConvolution_accGradParameters...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type   
not found: THNN_CudaHalfVolumetricFullConvolution_accGradParameters...me/alexander/torch/install/share/lua/5.1/nn/THNN.lua:108: NYI: call arg type  
model_opt   
{
  model_foveal_exclude : -1
  model_conv345_norm : true
  model_het : true
}
Warning: Failed to load function from bytecode: binary string: bad header in precompiled chunkWarning: Failed to load function from bytecode: binary string: bad header in precompiled chunkdev 1   1   
dev 2   2   
dev 3   1   
dev 4   2   
nn.Sequential {
  [architecture cut for brevity]
}
convert: ./data/annotations/instances_train2014.json --> .t7 [please be patient]    
converting: categories  
converting: annotations 
converting: images  
convert: building indices   
convert: complete [33.86 s] 
Loading proposals at    {
  1 : "/home/alexander/multipathnet/data/proposals/coco/sharpmask/train.t7"
  2 : "/home/alexander/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/alexander/multipathnet/data/proposals/coco/sharpmask/train.t7"
  2 : "/home/alexander/multipathnet/data/proposals/coco/sharpmask/val.t7"
}
Done loading proposals  
# proposal images   123287  
# dataset images    118287  
# images    123287  
nImages 118287  
/home/alexander/torch/install/bin/lua: ...alexander/torch/install/share/lua/5.1/trepl/init.lua:389: ./loaders/loader.lua:67: expected cdata for arg #1
stack traceback:
    [C]: in function 'error'
    ...alexander/torch/install/share/lua/5.1/trepl/init.lua:389: 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
    [C]: ?

Crashes on file_name = ffi.string(self.images.file_name[idx]), in loader.lua.

Could this be related in some way to it loading the same proposal data twice in a row?

Since the data in self.images.file_name[idx] appears to not be cdata -- I added a print statement to the start of the function and got a singular COCO_train2014_000000057870.jpg for each time it loaded the proposal data before failure -- I tried assigning it directly to file_name without the conversion. This saw the function getting past one image (does several thousand), but ran into the below error (I have torchnet).

[ same as above ]
[ model architecture ends here ]
}
Loading proposals at    {
  1 : "/home/alexander/multipathnet/data/proposals/coco/sharpmask/train.t7"
  2 : "/home/alexander/multipathnet/data/proposals/coco/sharpmask/val.t7"
}
Done loading proposals  
# proposal images   123287  
# dataset images    118287  
# images    123287  
nImages 118287  
FATAL THREAD PANIC: (read) ...alexander/torch/install/share/lua/5.1/torch/File.lua:343: unknown Torch class <package.torchnet>