Alm

This library implements common computer vision metrics in PyTorch. Algorithms are tested and can serve as an easy starting point to benchmark algorithms.

• Convex polygon intersection on the GPU using Sutherland-Hodgman algorithm
• Differentiable convex hull on the GPU
• Mean Average Precision (mAP) of polygons, Precision-Recall, IoU, GIoU
• Support for rotated bounding boxes

Polygon intersection on the GPU

Install

pip install git+https://github.com/janhuenermann/alm.git

Use Cases

Here are some tutorials to apply Alm to real world machine learning tasks:

• Compute Average Precision (for boxes and rotated boxes) in PyTorch Lightning