bossebandowski / xtcocoapi-bugfix

Extended COCO-API

Extended COCO API (xtcocotools)

News

[2022.12.27] Release xtcocotools v1.13. Fix int overflow & solve deprecation in numpy (replace np.float with np.float64).

[2022.04.10] Release xtcocotools v1.12. Fix bugs in APm and APl calculation.

[2022.02.23] Release xtcocotools v1.11. Add Windows/Mac support.

[2021.08.04] Release xtcocotools v1.10. Update installation dependencies.

[2021.07.22] Release xtcocotools v1.9. Merge some useful PRs from cocoapi.

[2021.05.19] Release xtcocotools v1.8. Fix CrowdPose evaluation.

[2021.03.22] Release xtcocotools v1.7. Support multi-part scores for COCO-WholeBody Dataset.

[2020.10.17] Release xtcocotools v1.6. Fix CrowdPose stats.

[2020.9.14] Release xtcocotools v1.5. Support COCO-WholeBody Dataset.

[2020.8.25] Release xtcocotools v1.0. Support COCO, AIChallenger, and CrowdPose Dataset.

Introduction

COCO has become a standard annotation format for the task of person keypoint detection, and is widely used for multiple datasets. Our Extended COCO API is developed based on @cocodataset/cocoapi.

We aim to provide a unified evaluation tools to support multiple human pose-related datasets, including COCO, COCO-WholeBody, CrowdPose, AI Challenger and so on.

xtcocotools has been used in MMPose framework.

We provide a simple demo_crowdpose to evaluate on CrowdPose dataset; demo_coco to evaluate on COCO dataset; and demo_coco_wholebody to evaluate on COCO-WholeBody dataset;

Requirements

• Python 3.7+ (Lower versions are not fully tested)

Installation

To install from pip:

pip install xtcocotools

To install from source:

pip install -r requirements.txt
python setup.py install