vdtk: Visual Description Evaluation Tools

This tool is designed to allow for a deep investigation of diversity in visual description datasets, and to help users understand their data at a token, n-gram, description, and dataset level.

Installation

To use this tool, you can easily pip install with pip install vdtk.

Data format

In order to prepare datasets to work with this tool, datasets must be formatted as JSON files with the following schema:

# List of samples in the dataset
[
    # JSON object for each sample
    {
        "_id": "string", # A string ID for each sample. This can help keep track of samples during use.
        "split": "string", # A string corresponding to the split of the data. Default splits are "train", "validate" and "test"
        "references": [
            # List of string references
            "reference 1...",
            "reference 2...",
        ],
        "candidates": [
            # List of string candidates (Optional)
            "candidate 1...",
            "candidate 2...",
        ],
        "media_path": "string", # (Optional) Path to the image/video (for image/video based metrics, recall experiemnts, etc.)
        "metadata": {} # Any JSON object. This field is not used by the toolkit at this time.
    }
]

Usage

After installation, the basic menu of commands can be accessed with vdtk --help. We make several experiments/tools available for use:

Additionally, several commands take multiple dataset JSONs, which can be used to compare different runs, or different datasets. Appending (:baseline) to any of the JSON file paths will treat this run as a baseline, and compute relative values and coloring accordingly (example: vdtk score cider-d ./baseline.json:baseline ./model.json).

For more details and options, see the --help command for any of the commands above. Note that some tools are relatively compute intensive. This toolkit will make use of a GPU if available and necessary, as well as a large number of CPU cores and RAM depending on the task.

Copyright 2021, Regents of the University of California

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.