This repository provides an unferenced image captioning metric from our ACL 2021 paper UMIC: An Unreferenced Metric for Image Captioning via Contrastive Learning.
Here, we provide the code to compute UMIC.

Usage (Updating the Descriptions)

Our code is based on UNITER. Therefore, please follow the install guideline for using Docker to load UNITER.
In the next few weeks, we try to release the version without using the docker.

1. Install Prerequisites

We used the Docker image provided by the official repo of UNITER. Using the guideline in the repo, please install the docker.

2. Download the Visual Features

For image captioning task, COCO dataset is widely used. To download the visual features for coco captions, just download the image features for coco validation splits using the following command.

Please refer to the offical repo of UNITER for downloading other visual features.

3. Pre-processing the Textual Features (Captions)

The format of textual feature file(python dictionary, json format) is as follows:
'cands' : [list of candidate captions]
'img_fs' : [list of image file names]

4. Running the Script

  1. Launching Docker
  1. Compute Score
python --data_type capeval1k \
                              --ckpt /storage/ \
                              --img_type \ coco_val2014 \


If you find this repo useful, please consider citing:

    title = "{UMIC}: An Unreferenced Metric for Image Captioning via Contrastive Learning",
    author = "Lee, Hwanhee  and
      Yoon, Seunghyun  and
      Dernoncourt, Franck  and
      Bui, Trung  and
      Jung, Kyomin",
    booktitle = "Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 2: Short Papers)",
    month = aug,
    year = "2021",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "",
    doi = "10.18653/v1/2021.acl-short.29",
    pages = "220--226",