Version 4.1.0 - Support for Keypoints, Boundaries, Occlusion, Superparts

Post-CVPR update to the Detail API for the PASCAL in Detail Challenge.

Changes:

• Added getKpts(), getBounds(), getOccl()
• Added superpart param to getMask()

Intended for use with the trainval_withkeypoints.json dataset.

See PythonAPI/ipynb/detailDemo.ipynb for the updated demo.

Version 4.0.1 - Bug fix

Small update to Version 4 of the Detail API for the PASCAL in Detail Challenge.

Changes:

• Fixed bug that caused loading for phases "training" and "validation" to only load part of the data
• Added phase parameter to getImgs()

Version 4 - Unified API

Fourth offficial release of the Detail API for the PASCAL in Detail challenge

This API version is intended for trainval_merged.json or later. Code for evaluating participants' algorithms is not yet provided and will be added in a future version.

Changes include;

• Dropped support for MATLAB API - see MATLAB tutorial for calling Python code from MATLAB
• Added segmentation and bounding box visualizations
• Refactored so that one unified interface is provided for working with semantic, instance, parts, detection, and objectness data. See detailDemo.ipynb

Version 3 - Support for Parts and Instance Segmentation

Third official release of the Detail API for the PASCAL in Detail challenge.

Both MATLAB and Python continue to be supported.

This API version is intended for use with the Parts dataset (trainval_parts.json). Code for evaluating participants' algorithms is not yet provided and will be added in future versions.

Changes include:

• Added getMask() function to Python API
• Updated MATLAB API

Release 1

Second official release of the Detail API for the PASCAL in Detail challenge.

Both MATLAB and Python continue to be supported.

This API version, like v2.1.0, is intended for use with the Preview dataset (trainval_preview1.json). Code for evaluating participants' algorithms is not yet provided and will be added in future versions.

Changes include:

• Added script for downloading datasets, download.py
• Added function showImgCat() to Python API

Preview

The first official release of the Detail API for the PASCAL in Detail challenge.

Both MATLAB and Python are supported.

This API version is intended for use with the Preview dataset. Code for evaluating participants' algorithms is not yet provided and will be added in future versions.