Panoptic Segmentation on Cityscapes val

Metric: mIoU (higher is better)

LeaderboardDataset

Loading chart...

Results

Hide extra data

Sort:

#	Model↕	mIoU▼	Extra Data	Paper	Date↕	Code
1	EfficientPS (Cityscapes-fine)	90.3	No	EfficientPS: Efficient Panoptic Segmentation	2020-04-05	Code
2	ViT-P (OneFormer, InternImage-H)	85.4	No	The Missing Point in Vision Transformers for Uni...	2025-05-26	Code
3	Panoptic-DeepLab (SWideRNet [1, 1, 4.5], Mapillary Vistas, multi-scale)	85.3	Yes	Scaling Wide Residual Networks for Panoptic Segm...	2020-11-23	-
4	OneFormer (ConvNeXt-L, single-scale, 512x1024, Mapillary Vistas-pretrained)	84.6	Yes	OneFormer: One Transformer to Rule Universal Ima...	2022-11-10	Code
5	Axial-DeepLab-XL (Mapillary Vistas, multi-scale)	84.6	Yes	Axial-DeepLab: Stand-Alone Axial-Attention for P...	2020-03-17	Code
6	Panoptic-DeepLab (SWideRNet [1, 1, 4.5], Mapillary Vistas, single-scale)	84.6	Yes	Scaling Wide Residual Networks for Panoptic Segm...	2020-11-23	-
7	OneFormer (ConvNeXt-XL, single-scale)	83.6	No	OneFormer: One Transformer to Rule Universal Ima...	2022-11-10	Code
8	kMaX-DeepLab (single-scale)	83.5	No	kMaX-DeepLab: k-means Mask Transformer	2022-07-08	Code
9	DiNAT-L (Mask2Former)	83.4	No	Dilated Neighborhood Attention Transformer	2022-09-29	Code
10	OneFormer (DiNAT-L, single-scale)	83.1	No	OneFormer: One Transformer to Rule Universal Ima...	2022-11-10	Code
11	OneFormer (ConvNeXt-L, single-scale)	83	No	OneFormer: One Transformer to Rule Universal Ima...	2022-11-10	Code
12	AFF-Base (single-scale, point-based Mask2Former)	83	No	AutoFocusFormer: Image Segmentation off the Grid	2023-04-24	Code
13	OneFormer (Swin-L, single-scale)	83	No	OneFormer: One Transformer to Rule Universal Ima...	2022-11-10	Code
14	Mask2Former (Swin-L)	82.9	No	Masked-attention Mask Transformer for Universal ...	2021-12-02	Code
15	AFF-Small (single-scale, point-based Mask2Former)	82.2	No	AutoFocusFormer: Image Segmentation off the Grid	2023-04-24	Code
16	EfficientPS	82.1	Yes	EfficientPS: Efficient Panoptic Segmentation	2020-04-05	Code
17	Panoptic-DeepLab (X71)	81.5	Yes	Panoptic-DeepLab: A Simple, Strong, and Fast Bas...	2019-11-22	Code
18	CMT-DeepLab (MaX-S, single-scale, IN-1K)	81.4	No	CMT-DeepLab: Clustering Mask Transformers for Pa...	2022-06-17	Code
19	Dynamically Instantiated Network (ResNet-101)	79.8	No	Weakly- and Semi-Supervised Panoptic Segmentation	2018-08-10	Code
20	COPS (ResNet-50)	79.3	No	Combinatorial Optimization for Panoptic Segmenta...	2021-06-06	Code
21	AdaptIS (ResNeXt-101)	79.2	No	AdaptIS: Adaptive Instance Selection Network	2019-09-17	-
22	UPSNet (ResNet-101, multiscale)	79.2	Yes	UPSNet: A Unified Panoptic Segmentation Network	2019-01-12	Code
23	TASCNet (ResNet-50, multi-scale)	78	Yes	Learning to Fuse Things and Stuff	2018-12-04	-
24	UPSNet (ResNet-101)	77.8	Yes	UPSNet: A Unified Panoptic Segmentation Network	2019-01-12	Code
25	TASCNet (ResNet-50)	77.8	Yes	Learning to Fuse Things and Stuff	2018-12-04	-
26	AdaptIS (ResNet-101)	77.2	No	AdaptIS: Adaptive Instance Selection Network	2019-09-17	-
27	Panoptic FPN (ResNet-101)	75.7	No	Panoptic Feature Pyramid Networks	2019-01-08	Code
28	AUNet (ResNet-101-FPN)	75.6	No	Attention-guided Unified Network for Panoptic Se...	2018-12-10	-
29	AdaptIS (ResNet-50)	75.3	No	AdaptIS: Adaptive Instance Selection Network	2019-09-17	-
30	UPSNet (ResNet-50)	75.2	No	UPSNet: A Unified Panoptic Segmentation Network	2019-01-12	Code

#1EfficientPS (Cityscapes-fine)SOTA
90.3
mIoU· 2020-04-05
EfficientPS: Efficient Panoptic Segmentation Code
#2ViT-P (OneFormer, InternImage-H)
85.4
mIoU· 2025-05-26
The Missing Point in Vision Transformers for Universal Image Segmentation Code
#3Panoptic-DeepLab (SWideRNet [1, 1, 4.5], Mapillary Vistas, multi-scale)
85.3
mIoU· Extra Data· 2020-11-23
Scaling Wide Residual Networks for Panoptic Segmentation
#4OneFormer (ConvNeXt-L, single-scale, 512x1024, Mapillary Vistas-pretrained)
84.6
mIoU· Extra Data· 2022-11-10
OneFormer: One Transformer to Rule Universal Image Segmentation Code
#5Axial-DeepLab-XL (Mapillary Vistas, multi-scale)SOTA
84.6
mIoU· Extra Data· 2020-03-17
Axial-DeepLab: Stand-Alone Axial-Attention for Panoptic Segmentation Code
#6Panoptic-DeepLab (SWideRNet [1, 1, 4.5], Mapillary Vistas, single-scale)
84.6
mIoU· Extra Data· 2020-11-23
Scaling Wide Residual Networks for Panoptic Segmentation
#7OneFormer (ConvNeXt-XL, single-scale)
83.6
mIoU· 2022-11-10
OneFormer: One Transformer to Rule Universal Image Segmentation Code
#8kMaX-DeepLab (single-scale)
83.5
mIoU· 2022-07-08
kMaX-DeepLab: k-means Mask Transformer Code
#9DiNAT-L (Mask2Former)
83.4
mIoU· 2022-09-29
Dilated Neighborhood Attention Transformer Code
#10OneFormer (DiNAT-L, single-scale)
83.1
mIoU· 2022-11-10
OneFormer: One Transformer to Rule Universal Image Segmentation Code
#11OneFormer (ConvNeXt-L, single-scale)
83
mIoU· 2022-11-10
OneFormer: One Transformer to Rule Universal Image Segmentation Code
#12AFF-Base (single-scale, point-based Mask2Former)
83
mIoU· 2023-04-24
AutoFocusFormer: Image Segmentation off the Grid Code
#13OneFormer (Swin-L, single-scale)
83
mIoU· 2022-11-10
OneFormer: One Transformer to Rule Universal Image Segmentation Code
#14Mask2Former (Swin-L)
82.9
mIoU· 2021-12-02
Masked-attention Mask Transformer for Universal Image Segmentation Code
#15AFF-Small (single-scale, point-based Mask2Former)
82.2
mIoU· 2023-04-24
AutoFocusFormer: Image Segmentation off the Grid Code
#16EfficientPS
82.1
mIoU· Extra Data· 2020-04-05
EfficientPS: Efficient Panoptic Segmentation Code
#17Panoptic-DeepLab (X71)SOTA
81.5
mIoU· Extra Data· 2019-11-22
Panoptic-DeepLab: A Simple, Strong, and Fast Baseline for Bottom-Up Panoptic Segmentation Code
#18CMT-DeepLab (MaX-S, single-scale, IN-1K)
81.4
mIoU· 2022-06-17
CMT-DeepLab: Clustering Mask Transformers for Panoptic Segmentation Code
#19Dynamically Instantiated Network (ResNet-101)SOTA
79.8
mIoU· 2018-08-10
Weakly- and Semi-Supervised Panoptic Segmentation Code
#20COPS (ResNet-50)
79.3
mIoU· 2021-06-06
Combinatorial Optimization for Panoptic Segmentation: A Fully Differentiable Approach Code
#21AdaptIS (ResNeXt-101)
79.2
mIoU· 2019-09-17
AdaptIS: Adaptive Instance Selection Network
#22UPSNet (ResNet-101, multiscale)
79.2
mIoU· Extra Data· 2019-01-12
UPSNet: A Unified Panoptic Segmentation Network Code
#23TASCNet (ResNet-50, multi-scale)
78
mIoU· Extra Data· 2018-12-04
Learning to Fuse Things and Stuff
#24UPSNet (ResNet-101)
77.8
mIoU· Extra Data· 2019-01-12
UPSNet: A Unified Panoptic Segmentation Network Code
#25TASCNet (ResNet-50)
77.8
mIoU· Extra Data· 2018-12-04
Learning to Fuse Things and Stuff
#26AdaptIS (ResNet-101)
77.2
mIoU· 2019-09-17
AdaptIS: Adaptive Instance Selection Network
#27Panoptic FPN (ResNet-101)
75.7
mIoU· 2019-01-08
Panoptic Feature Pyramid Networks Code
#28AUNet (ResNet-101-FPN)
75.6
mIoU· 2018-12-10
Attention-guided Unified Network for Panoptic Segmentation
#29AdaptIS (ResNet-50)
75.3
mIoU· 2019-09-17
AdaptIS: Adaptive Instance Selection Network
#30UPSNet (ResNet-50)
75.2
mIoU· 2019-01-12
UPSNet: A Unified Panoptic Segmentation Network Code