AI Paper Atlas

Research map / Benchmarks / ADE20K

ADE20K leaderboard

357 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on ADE20K. Best mIoU: 73.92 (LH-180, NeurIPS 2024).

Numbers are as each paper reports them, read by AI from its results tables and not checked by hand. Papers can differ in setting (training data, model size, extra data); the topic column shows what each paper is about. Open the paper before you compare or quote a number.

ADE20K · mIoU

223 papers, each with its best reported number for its own method; higher is better.

#mIoU ↑MethodPaperTopic
173.92LH-180
ViT-B/16
LookHere: Vision Transformers with Directed Attention Generalize and Extrapolate
NeurIPS 2024 · Table: Semantic Segmentation
Attention and transformers
272.94Spline-GCN
Cityscapes-trained ResNet-50
Fast Interactive Object Annotation With Curve-GCN
CVPR 2019 · Table 6
CNN features
368.4MQ-Former
Swin Transformer
Scaling up Image Segmentation across Data and Tasks
CVPR 2025 · Table 3
Open-set and few-shot detection
462CDRA-SPT
DINOv2
Sensitivity-Aware Efficient Fine-Tuning via Compact Dynamic-Rank Adaptation
CVPR 2025 · Table 3
Pruning and efficient tuning
561.4FD-SwinV2-G
SwinV2-G
Improving CLIP Fine-tuning Performance
ICCV 2023 · Table 9
CNN features
659.9SimMIM
SwinV2-G
SimMIM: A Simple Framework for Masked Image Modeling
CVPR 2022 · text
Attention and transformers
758.6Depth Anything V2 Large
Depth Anything V2
Depth Anything V2
NeurIPS 2024 · Table: Semantic segmentation
Recognition and long-tail
857.9BEiT-L+
ViT-Large
BEiT: BERT Pre-Training of Image Transformers
ICLR 2022 · Large-scale ADE20K table
Human and embodied modeling
956.5Argus
DINOv2 ViT-Base
Argus: A Compact and Versatile Foundation Model for Vision
CVPR 2025 · Table 1
Human and embodied modeling
1056.4Mask2Former + Swin-L (synthetic pre-train, real fine-tune)
Swin-L
FreeMask: Synthetic Images with Dense Annotations Make Stronger Segmentation Models
NeurIPS 2023 · Table: pre-training on ADE20K
Open-set and few-shot detection
213 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ADE20K · mIoU (val)

125 papers, each with its best reported number for its own method; higher is better.

#mIoU ↑MethodPaperTopic
158.22Mask2Former+IDRNet
Swin-Large
IDRNet: Intervention-Driven Relation Network for Semantic Segmentation
NeurIPS 2023 · Table SOTA
CNN features
256.7GMMSeg
Mask2Former Swin-L
GMMSeg: Gaussian Mixture based Generative Semantic Segmentation Models
NeurIPS 2022 · Appendix Table
CNN features
356.6BEiT-L + UPerNet + CTS
BEiT-L
Content-Aware Token Sharing for Efficient Semantic Segmentation With Vision Transformers
CVPR 2023 · Table: Application to state-of
CNN features
456.5Mask2Former + DejaVu
Mask2Former
DejaVu: Conditional Regenerative Learning To Enhance Dense Prediction
CVPR 2023 · Table 3
CNN features
554.3VOLO-D5VOLO: Vision Outlooker for Visual Recognition
TPAMI 2022 · ADE20K table
Attention and transformers
654.3UperNet + K-Net (640)
Swin-L
K-Net: Towards Unified Image Segmentation
NeurIPS 2021 · Table 3b
CNN features
754.2LRFormer-L
ImageNet-22K pretrained
Low-Resolution Self-Attention for Semantic Segmentation
TPAMI 2025 · Table 2
CNN features
854.2TSG
Swin-Large
Transformer Scale Gate for Semantic Segmentation
CVPR 2023 · Table 2
CNN features
953.8MambaMLP-L (MFD + SLWS)
MambaMLP-L
Stochastic Layer-Wise Shuffle for Improving Vision Mamba Training
ICML 2025 · Semantic segmentation results
Attention and transformers
1053.7DLPL (SegFormer MiT-B5)
SegFormer MiT-B5
Discrete Latent Perspective Learning for Segmentation and Detection
ICML 2024 · Auto-Driving and Daily photos
Open-set and few-shot detection
115 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ADE20K · FID

26 papers, each with its best reported number for its own method; lower is better.

16 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ADE20K · mIoU (ms) (val)

12 papers, each with its best reported number for its own method; higher is better.

#mIoU (ms) ↑MethodPaperTopic
157.7Mask2FormerMasked-Attention Mask Transformer for Universal Image Segmentation
CVPR 2022 · Table 3
CNN features
255.6MaskFormer
Swin-L
Per-Pixel Classification is Not All You Need for Semantic Segmentation
NeurIPS 2021 · Table 1
CNN features
355.2ECENet
SwinT-Large
Boosting Semantic Segmentation from the Perspective of Explicit Class Embeddings
ICCV 2023 · Table 1
CNN features
455.2SegViT
ViT-Large
SegViT: Semantic Segmentation with Plain Vision Transformers
NeurIPS 2022 · Table 1
CNN features
554.3DEPICT-SA
ViT-L
Rethinking Decoders for Transformer-based Semantic Segmentation: A Compression Perspective
NeurIPS 2024 · Table: Comparison on ADE20K va
CNN features
653.4RegProxy-L/16
ViT-L/16
Semantic Segmentation by Early Region Proxy
CVPR 2022 · Main Results table
CNN features
752.1SegNeXt-L
MSCAN-L
SegNeXt: Rethinking Convolutional Attention Design for Semantic Segmentation
NeurIPS 2022 · Table: ADE20K, Cityscapes and
CNN features
851.8PPL + Semantic FPN
ViT-B
Probabilistic Prompt Learning for Dense Prediction
CVPR 2023 · Table: Quantitative results of
CLIP and zero-shot
951.8LV-ViT-L + UperNetAll Tokens Matter: Token Labeling for Training Better Vision Transformers
NeurIPS 2021 · Table 6
Attention and transformers
1051.3DenseCLIP + Semantic FPN
CLIP ViT-B
DenseCLIP: Language-Guided Dense Prediction With Context-Aware Prompting
CVPR 2022 · Table 1
Open-set and few-shot detection
2 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

Not affiliated with any conference, journal or the benchmark's authors. Corrections: contact@latenticlabs.com.