MODELS.md

Models

Here, we provide model weights and config files for the main models from our CVPR 2024 paper.

Pascal-PP

Method	Backbone	Pre-training	PartPQ_Pt	PartPQ_NoPt	PartPQ_All	PartSQ_Pt	PQ	config	model
TAPPS	R50	ImageNet1K	59.6	39.4	44.6	74.3	47.1	config	model
TAPPS	R50	COCO	67.2	50.4	54.7	75.1	57.7	config	model
TAPPS	Swin-B	COCO	72.2	56.3	60.4	78.1	63.0	config	model

Pascal-PP-107

Method	Backbone	Pre-training	PartPQ_Pt	PartPQ_NoPt	PartPQ_All	PartSQ_Pt	PQ	config	model
TAPPS	R50	ImageNet1K	49.8	39.3	42.0	61.7	47.2	config	model
TAPPS	R50	COCO	55.4	50.0	51.4	62.1	57.4	config	model

Cityscapes-PP

Method	Backbone	Pre-training	PartPQ_Pt	PartPQ_NoPt	PartPQ_All	PartSQ_Pt	PQ	config	model
TAPPS	R50	ImageNet1K	48.7	63.1	59.3	66.8	62.4	config	model
TAPPS	R50	COCO	48.9	65.7	61.3	66.9	64.4	config	model
TAPPS	Swin-B	COCO	53.0	69.0	64.8	68.0	68.0	config	model