tf-convex-quad-iou

Convex Quad Intersection over Union (IoU) computation for tensorflow.

This repository is a rewrite of the algorithm in https://github.com/NVIDIA/retinanet-examples to compute intersections of convex quads. The algorithm have been lightly modified to have a better handling of coincident corners.

Installation

pip install tf-convex-quad-iou-atuleu

Installing from source

Requirements

• tensorflow >= 2.8.0
• bazelisk

Steps by steps

If you want to compile both CPU and GPU custom operation, please set the global variable TF_NEED_CUDA to 1

git clone https://github.com/atuleu/tf-convex-quad-iou.git
cd tf-convex-quad-iou

export TF_NEED_CUDA="1" # do not export if you do not need cuda support
python3 configure.py

bazel build build_pip_pkg
bazel-bin/build_pip_pkg artifacts

pip install artifacts/tf_convex_quad_iou_atuleu*.whl