Awesome

COCO-CN output inspector

This is a lightweight tool to inspect the results of different image description systems on the COCO-CN test set. It automatically downloads the images as you browse through the results. This saves you from having to download all the images in the dataset. I've also provided a script to download the images, if you need to work offline.

Requirements

• Python 3.6
• Flask 0.12

Other versions may work, but remain untested. I recommend downloading Anaconda.

Screenshot

Here's a preview of the tool. There's a simple navigation menu (previous, next), with the image below, followed by the human-generated descriptions, the descriptions generated by the systems, and the evaluation metrics.

How to run this tool

To run the tool, just use python resultsviewer.py on the command line. If you want to download the images first, run python download_images.py and wait for the script to complete. This may take a long time.

Contact

• Tool: Emiel van Miltenburg, Tilburg University
• Data: Xirong Li, Renmin University of China. (See their GitHub here)

Reference

If you find this repository useful, please

• Link back to it in your work.
• Cite the original paper by Li et al. See:

@article{DBLP:journals/corr/abs-1805-08661,
  author    = {Xirong Li and
               Xiaoxu Wang and
               Chaoxi Xu and
               Weiyu Lan and
               Qijie Wei and
               Gang Yang and
               Jieping Xu},
  title     = {{COCO-CN} for Cross-Lingual Image Tagging, Captioning and Retrieval},
  journal   = {CoRR},
  volume    = {abs/1805.08661},
  year      = {2018},
  url       = {http://arxiv.org/abs/1805.08661},
  archivePrefix = {arXiv},
  eprint    = {1805.08661},
  timestamp = {Mon, 13 Aug 2018 16:48:10 +0200},
  biburl    = {https://dblp.org/rec/bib/journals/corr/abs-1805-08661},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}