evaluate_model error of 'bad argument'

[sujing1]

Hello, I get error when running 'evaluate_model.lua'. It appears with both VG 1.0 and 1.2: THNN.lua:110: bad argument #8 to 'v' (cannot convert 'nil' to 'int64_t')

Any suggestions are highly appreciated!

Steps

1. Necessary packages are installed as from the Installation guideline

2. Image preprocessing succeeds with python preprocess.py --region_data ~/data/VG2015/region_descriptions.json --image_dir ~/data/VG2015/VG_100K --image_size 720 --split_json info/densecap_splits.json --json_output ~/data/VG2015/VG_100k_sorted.json --h5_output ~/data/VG2015/VG_100k_sorted.hdf5 --num_workers 6

3. Error comes with th evaluate_model.lua -checkpoint data/models/densecap/densecap-pretrained-vgg16.t7 -split val -gpu 0 -use_cudnn 0 -num_proposals 300 -data_h5 ~/data/VG2015/VG_100k_sorted.hdf5 -data_json ~/data/VG2015/VG_100k_sorted.json

Loaded model Using dtype "torch.CudaTensor" DataLoader loading json file: /home/sujing/data/VG2015/VG_100k_sorted.json DataLoader loading h5 file: /home/sujing/data/VG2015/VG_100k_sorted.hdf5 reading box_to_img reading boxes reading image_heights reading image_widths reading img_to_first_box reading img_to_last_box reading labels reading lengths reading original_heights reading original_widths reading split assigned 77396/5000/5000 images to train/val/test. initialized DataLoader: #images: 87396, #regions: 4062806, sequence max length: 15 using split 1 /home/sujing/workspace/torch/install/bin/luajit: ...g/workspace/torch/install/share/lua/5.1/nn/Container.lua:67: In 3 module of nn.Sequential: ...sujing/workspace/torch/install/share/lua/5.1/nn/THNN.lua:110: bad argument #8 to 'v' (cannot convert 'nil' to 'int64_t') stack traceback: [C]: in function 'v' ...sujing/workspace/torch/install/share/lua/5.1/nn/THNN.lua:110: in function 'ClassNLLCriterion_updateOutput' ...ace/torch/install/share/lua/5.1/nn/ClassNLLCriterion.lua:41: in function 'updateOutput' ./densecap/modules/OurCrossEntropyCriterion.lua:19: in function 'forward' ./densecap/LocalizationLayer.lua:466: in function 'f' ./densecap/LocalizationLayer.lua:228: in function 'timeit' ./densecap/LocalizationLayer.lua:455: in function <./densecap/LocalizationLayer.lua:383> [C]: in function 'xpcall' ...g/workspace/torch/install/share/lua/5.1/nn/Container.lua:63: in function 'rethrowErrors' .../workspace/torch/install/share/lua/5.1/nn/Sequential.lua:44: in function 'forward' ./densecap/DenseCapModel.lua:253: in function 'forward' ./densecap/DenseCapModel.lua:406: in function 'forward_backward' ./eval/eval_utils.lua:58: in function 'eval_split' evaluate_model.lua:63: in main chunk [C]: in function 'dofile' ...pace/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:150: in main chunk [C]: at 0x00405d50

[lixyresearch]

I have the same problem. How to fix it?

[lixyresearch]

I have solved it. I have trained a new model, and the new one works well.

[doiken23]

Could you reproduce the score of the original Densecap parper? I have the same problem and consider to train a new model from scratch.