search

naver-ai / pcme

Official Pytorch implementation of "Probabilistic Cross-Modal Embedding" (CVPR 2021)
Other
122 stars 17 forks source link

.t7 caption files #1

Closed YangYang closed 3 years ago

YangYang commented 3 years ago

I'm already following the readme to download the CUB dataset, but captions in there are not suitable for this code. Because of the captions file type is .t7 not .txt.

SanghyukChun commented 3 years ago

@YangYang Hi, thanks for your question. Could you provide a more detailed log? I have confirmed that the following command works for my environment

python train_cub.py ./config/cub/pcme_cub.yaml --dataset_root <dataset_root> --caption_root <caption_root>/cvpr2016_cub/text_c10

Note that, cvpr2016_cub/text_c10 looks like this:

├── 001.Black_footed_Albatross.t7                       # t7 file you mentioned (not used by my code)
├── 001.Black_footed_Albatross                          # per-instance caption data used by my code
│   ├── Black_Footed_Albatross_0001_796111.h5           # not used by my code
│   ├── Black_Footed_Albatross_0001_796111.txt          # actual txt file loaded by my code
│   ├── Black_Footed_Albatross_0002_55.h5
│   ├── Black_Footed_Albatross_0002_55.txt
...
YangYang commented 3 years ago

Yes, the composition of cvpr2016_cub/text_c10 is exactly what you said. The reason I can't run this code is the image name doesn't match the caption name. According to the link you gave to the image, I could get: images/001.Black_footed_Albatross

  ├── Black_footed_Albatross_0001_2950163169.jpg                    
  ├── Black_footed_Albatross_0002_2293084168.jpg 
 ...

However, in the caption, the cvpr2016_cub/text_c10/001.Black_footed_Albatross is contains:

  ├── Black_Footed_Albatross_0001_796111.txt
  ├── Black_Footed_Albatross_0002_55.txt
 ...
sawant-nidhish commented 3 years ago

@SanghyukChun Sir, Firstly Thank You for the briliant work! Even I am facing the same problem as @YangYang.

Yes, the composition of cvpr2016_cub/text_c10 is exactly what you said. The reason I can't run this code is the image name doesn't match the caption name. According to the link you gave to the image, I could get: images/001.Black_footed_Albatross

  ├── Black_footed_Albatross_0001_2950163169.jpg                    
  ├── Black_footed_Albatross_0002_2293084168.jpg 
 ...

However, in the caption, the cvpr2016_cub/text_c10/001.Black_footed_Albatross is contains:

  ├── Black_Footed_Albatross_0001_796111.txt
  ├── Black_Footed_Albatross_0002_55.txt
 ...
SanghyukChun commented 3 years ago

@YangYang @sawant-nidhish Oh, I see. Thanks for your detailed reports! I will take a look into the issue, and revise the code as soon as possible.

YangYang commented 3 years ago

@SanghyukChun Thank you for your attention! I will wait patiently for your reply.

SanghyukChun commented 3 years ago

@YangYang @sawant-nidhish Aha, I found that the link for the CUB dataset is wrong. I used CUB-200-2011, the extended version, but the link is the 2010 version. Please re-download the dataset in http://www.vision.caltech.edu/visipedia/CUB-200-2011.html I just revised the README document

Thanks for your questions!