google-research/deeplab2
DeepLab2 is a Google TensorFlow library providing state-of-the-art models for dense pixel labeling tasks like semantic and panoptic segmentation.
Velocity · 7d
+0.6
★ / day
Trend
→steady
star history
DeepLab2 provides a unified TensorFlow codebase for deep labeling, which refers to solving computer vision problems by assigning predicted values to each pixel in an image using deep neural networks. The library includes implementations of recent research models such as kMaX-DeepLab, ViP-DeepLab, and MOAT for tasks including semantic segmentation, instance segmentation, panoptic segmentation, depth estimation, and video panoptic segmentation.