oarriaga/paz

Built on TensorFlow 2.0, OpenCV, and NumPy, PAZ provides a hierarchical API spanning high-level application pipelines down to low-level processor composition, enabling flexible computer vision workflows from object detection to 6D pose estimation. The mid-level `SequentialProcessor` abstraction allows composing reusable perception chains (data augmentation, preprocessing, inference) without boilerplate, while pre-trained models cover diverse tasks including probabilistic keypoint estimation, emotion classification, and semantic segmentation. All included models support retraining on custom datasets, making it suitable for autonomous systems requiring modular, customizable perception stacks.

702 stars.