panoptic_deeplab

Panoptic-Deeplab: A Simple, Strong, and Fast Baseline for Bottom-Up Panoptic Segmentation

Reference

Cheng, Bowen, et al. "Panoptic-deeplab: A simple, strong, and fast baseline for bottom-up panoptic segmentation." Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2020.

Performance

Cityscapes

Model	Backbone	Resolution	Training Iters	PQ	mIoU	mAP50	Links
Panoptic-DeepLab	ResNet50-vd	1025x513	90k	60.32%	46.34%	79.68%	config | model

• The models were trained using 8 GPUs.
• We observed better performance by using the input resolution of 1025x513 instead of 1024x512 and setting align_corners to True during training.