Cocotools

The cocotools crate provides tools to load, manipulate, convert and visualize COCO format datasets.

This crate aims to provide similar functionalities to the python pycocotools package / cocoapi with additionnal utilities such as conversion between dataset formats. It also aims to have a better documentation and a more readable implementation.

API Usage example

use std::path::PathBuf;
use cocotools::COCO;

let annotations_file_path = PathBuf::from("../data_samples/coco_25k/annotations.json");
let dataset = COCO::try_from(&annotations_file_path)?;
let file_name = dataset.get_img(17627)?.file_name;

Program Usage

cargo run -- visualize  ../data_samples/coco_25k/annotations.json ../data_samples/coco_25k/images -s 000000017627
cargo run -- convert-segmentation ../data_samples/coco_25k/annotations.json rle -o annotations_rle.json

Planned features

• Add support for keypoint detection format.
• Add conversion from/to PascalVOC format.
• Add conversion from/to SOLO format.
• Add validation of the data when loading it, for example check that sum(rle) == nb pixels in the image (behind a crate feature flags ?)
• Use rayon when loading/converting the data ?