Panoptic Segmentation on ADE20K val

Metric: mIoU (higher is better)

LeaderboardDataset

Loading chart...

Results

Hide extra data

Sort:

#	Model↕	mIoU▼	Extra Data	Paper	Date↕	Code
1	OneFormer (InternImage-H, emb_dim=256, single-scale, 896x896)	60.4	No	OneFormer: One Transformer to Rule Universal Ima...	2022-11-10	Code
2	X-Decoder (Davit-d5, Deform, single-scale, 1280x1280)	59.1	Yes	Generalized Decoding for Pixel, Image, and Langu...	2022-12-21	Code
3	OneFormer (DiNAT-L, single-scale, 1280x1280, COCO-Pretrain)	58.9	Yes	OneFormer: One Transformer to Rule Universal Ima...	2022-11-10	Code
4	OneFormer (DiNAT-L, single-scale, 1280x1280)	58.3	No	OneFormer: One Transformer to Rule Universal Ima...	2022-11-10	Code
5	OneFormer (DiNAT-L, single-scale, 640x640)	58.3	No	OneFormer: One Transformer to Rule Universal Ima...	2022-11-10	Code
6	X-Decoder (L)	58.1	Yes	Generalized Decoding for Pixel, Image, and Langu...	2022-12-21	Code
7	OneFormer (ConvNeXt-XL, single-scale, 640x640)	57.4	No	OneFormer: One Transformer to Rule Universal Ima...	2022-11-10	Code
8	OneFormer (Swin-L, single-scale, 1280x1280)	57	No	OneFormer: One Transformer to Rule Universal Ima...	2022-11-10	Code
9	OneFormer (Swin-L, single-scale, 640x640)	57	No	OneFormer: One Transformer to Rule Universal Ima...	2022-11-10	Code
10	OneFormer (ConvNeXt-L, single-scale, 640x640)	56.6	No	OneFormer: One Transformer to Rule Universal Ima...	2022-11-10	Code
11	DiNAT-L (Mask2Former, 640x640)	56.3	No	Dilated Neighborhood Attention Transformer	2022-09-29	Code
12	Mask2Former (Swin-L + FAPN, 640x640)	55.4	No	Masked-attention Mask Transformer for Universal ...	2021-12-02	Code
13	kMaX-DeepLab (ConvNeXt-L, single-scale, 1281x1281)	55.2	No	kMaX-DeepLab: k-means Mask Transformer	2022-07-08	Code
14	kMaX-DeepLab (ConvNeXt-L, single-scale, 641x641)	54.8	No	kMaX-DeepLab: k-means Mask Transformer	2022-07-08	Code
15	Mask2Former (Swin-L)	54.5	No	Masked-attention Mask Transformer for Universal ...	2021-12-02	Code
16	Panoptic-DeepLab (SwideRNet)	50	No	Masked-attention Mask Transformer for Universal ...	2021-12-02	Code
17	Mask2Former (ResNet-50, 640x640)	46.1	No	Masked-attention Mask Transformer for Universal ...	2021-12-02	Code
18	kMaX-DeepLab (ResNet50, single-scale, 1281x1281)	45.3	No	kMaX-DeepLab: k-means Mask Transformer	2022-07-08	Code
19	kMaX-DeepLab (ResNet50, single-scale, 641x641)	45	No	kMaX-DeepLab: k-means Mask Transformer	2022-07-08	Code

#1OneFormer (InternImage-H, emb_dim=256, single-scale, 896x896)SOTA
60.4
mIoU· 2022-11-10
OneFormer: One Transformer to Rule Universal Image Segmentation Code
#2X-Decoder (Davit-d5, Deform, single-scale, 1280x1280)
59.1
mIoU· Extra Data· 2022-12-21
Generalized Decoding for Pixel, Image, and Language Code
#3OneFormer (DiNAT-L, single-scale, 1280x1280, COCO-Pretrain)
58.9
mIoU· Extra Data· 2022-11-10
OneFormer: One Transformer to Rule Universal Image Segmentation Code
#4OneFormer (DiNAT-L, single-scale, 1280x1280)
58.3
mIoU· 2022-11-10
OneFormer: One Transformer to Rule Universal Image Segmentation Code
#5OneFormer (DiNAT-L, single-scale, 640x640)
58.3
mIoU· 2022-11-10
OneFormer: One Transformer to Rule Universal Image Segmentation Code
#6X-Decoder (L)
58.1
mIoU· Extra Data· 2022-12-21
Generalized Decoding for Pixel, Image, and Language Code
#7OneFormer (ConvNeXt-XL, single-scale, 640x640)
57.4
mIoU· 2022-11-10
OneFormer: One Transformer to Rule Universal Image Segmentation Code
#8OneFormer (Swin-L, single-scale, 1280x1280)
57
mIoU· 2022-11-10
OneFormer: One Transformer to Rule Universal Image Segmentation Code
#9OneFormer (Swin-L, single-scale, 640x640)
57
mIoU· 2022-11-10
OneFormer: One Transformer to Rule Universal Image Segmentation Code
#10OneFormer (ConvNeXt-L, single-scale, 640x640)
56.6
mIoU· 2022-11-10
OneFormer: One Transformer to Rule Universal Image Segmentation Code
#11DiNAT-L (Mask2Former, 640x640)SOTA
56.3
mIoU· 2022-09-29
Dilated Neighborhood Attention Transformer Code
#12Mask2Former (Swin-L + FAPN, 640x640)SOTA
55.4
mIoU· 2021-12-02
Masked-attention Mask Transformer for Universal Image Segmentation Code
#13kMaX-DeepLab (ConvNeXt-L, single-scale, 1281x1281)
55.2
mIoU· 2022-07-08
kMaX-DeepLab: k-means Mask Transformer Code
#14kMaX-DeepLab (ConvNeXt-L, single-scale, 641x641)
54.8
mIoU· 2022-07-08
kMaX-DeepLab: k-means Mask Transformer Code
#15Mask2Former (Swin-L)
54.5
mIoU· 2021-12-02
Masked-attention Mask Transformer for Universal Image Segmentation Code
#16Panoptic-DeepLab (SwideRNet)
50
mIoU· 2021-12-02
Masked-attention Mask Transformer for Universal Image Segmentation Code
#17Mask2Former (ResNet-50, 640x640)
46.1
mIoU· 2021-12-02
Masked-attention Mask Transformer for Universal Image Segmentation Code
#18kMaX-DeepLab (ResNet50, single-scale, 1281x1281)
45.3
mIoU· 2022-07-08
kMaX-DeepLab: k-means Mask Transformer Code
#19kMaX-DeepLab (ResNet50, single-scale, 641x641)
45
mIoU· 2022-07-08
kMaX-DeepLab: k-means Mask Transformer Code