AI Paper Atlas

Research map / Benchmarks / Cityscapes

Cityscapes leaderboard

387 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on Cityscapes. Best mIoU: 86.12 (SeeThroughNet, CVPR 2022).

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.

Cityscapes · mIoU

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

#mIoU ↑MethodPaperTopic
186.12SeeThroughNet
HRNetV2-W48
SeeThroughNet: Resurrection of Auxiliary Loss by Preserving Class Probability Information
CVPR 2022 · Table 6
CNN features
285.8Seg-VARSeg-VAR:Image Segmentation with Visual Autoregressive Modeling
NeurIPS 2025 · Table: Results on Image Segmen
Autoregressive and latent generation
385.6HMS + InverseForm
HRNet48 HMS
InverseForm: A Loss Function for Structured Boundary-Aware Segmentation
CVPR 2021 · Table 4
CNN features
485.3Ours (warp-refine labels)Warp-Refine Propagation: Semi-Supervised Auto-Labeling via Cycle-Consistency
ICCV 2021 · Table 2
Semi-supervised learning
585.1DAS + ResNeSt
ResNet-200
Your Image is My Video: Reshaping the Receptive Field via Image-To-Video Differentiable AutoAugmentation and Fusion
CVPR 2024 · Table 3
Video segmentation and tracking
684.5HRNet + OCR + SegFix
HRNet+OCR
SegFix: Model-Agnostic Boundary Refinement for Segmentation
ECCV 2020 · Table (test mIoU)
CNN features
784.3DCNAS + ASPPDCNAS: Densely Connected Neural Architecture Search for Semantic Image Segmentation
CVPR 2021 · Table: Performance on Cityscap
Neural architecture search
883.8SegMAN-BSegMAN: Omni-scale Context Modeling with State Space Models and Local Attention for Semantic Segmentation
CVPR 2025 · Table 2
CNN features
983.6FLANet
HRNetW48
Fully Attentional Network for Semantic Segmentation
AAAI 2022 · Table 1
CNN features
1083.6DeepLabv3+ with Euclidean Frank-Wolfe CRF
DeepLabv3+
Regularized Frank-Wolfe for Dense CRFs: Generalizing Mean Field and Beyond
NeurIPS 2021 · Table: Performance on the test
Kernels and regression theory
222 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

Cityscapes · mIoU (val)

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

#mIoU ↑MethodPaperTopic
187.1HMS + DejaVu
HMS
DejaVu: Conditional Regenerative Learning To Enhance Dense Prediction
CVPR 2023 · Table 1
CNN features
287HMS + InverseForm
HRNet48 HMS
InverseForm: A Loss Function for Structured Boundary-Aware Segmentation
CVPR 2021 · Table 4
CNN features
385.1DLPL (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
484.04HiPoSeg
DeepLabV3+ ResNet-101
Hierarchical Prototype Learning for Semantic Segmentation
ICLR 2026 · Table 1
Open-set and few-shot detection
584SegFormer-B5
MiT-B5
SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers
NeurIPS 2021 · Table: Comparison to state of
CNN features
683.52Swin-L + UperNet (efficient attn) + SegDeformer (IN-22K)
Swin-L
A Transformer-Based Decoder for Semantic Segmentation with Multi-level Context Mining
ECCV 2022 · Table 3
CNN features
783.5kMaX-DeepLab
ConvNeXt-L (ImageNet-22K)
k-Means Mask Transformer
ECCV 2022 · Table 2
CNN features
883.2LRFormer-LLow-Resolution Self-Attention for Semantic Segmentation
TPAMI 2025 · Table 4
CNN features
983.2UFP+GraphFPNGraphFPN: Graph Feature Pyramid Network for Object Detection
ICCV 2021 · Table (appendix)
CNN features
1083.16HRViT-b3Multi-Scale High-Resolution Vision Transformer for Semantic Segmentation
CVPR 2022 · Cityscapes comparison table
CNN features
137 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

Cityscapes · PQ (val)

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

#PQ ↑MethodPaperTopic
168.4kMaX-DeepLab
ConvNeXt-L (ImageNet-22K)
k-Means Mask Transformer
ECCV 2022 · Table 2
CNN features
267.2OneFormer
Swin-L
OneFormer: One Transformer To Rule Universal Image Segmentation
CVPR 2023 · Table 2
CNN features
367Panoptic-DeepLab
Xception-71
Panoptic-DeepLab: A Simple, Strong, and Fast Baseline for Bottom-Up Panoptic Segmentation
CVPR 2020 · Table: Cityscapes val set
CNN features
466.7Crop + CABB + ISUS
HRNetV2-W48+
Improving Panoptic Segmentation at All Scales
CVPR 2021 · Table 1
CNN features
565.9Panoptic FCN*
SwinT-L-FPN
Fully Convolutional Networks for Panoptic Segmentation with Point-based Supervision
TPAMI 2022 · Table 15
Weakly supervised detection and segmentation
665.4ReMaX R50
ResNet-50
ReMaX: Relaxing for Better Training on Efficient Panoptic Segmentation
NeurIPS 2023 · Table 7
CNN features
764.6CMT-DeepLab-SCMT-DeepLab: Clustering Mask Transformers for Panoptic Segmentation
CVPR 2022 · Cityscapes results table
CNN features
864.1Panoptic-DeepLab + PISR
Panoptic-DeepLab, HRNet-w48
Panoptic, Instance and Semantic Relations: A Relational Context Encoder To Enhance Panoptic Segmentation
CVPR 2022 · Table 1
CNN features
964.1PanopticDepth + FSF
R-50
PanopticDepth: A Unified Framework for Depth-Aware Panoptic Segmentation
CVPR 2022 · Table 1
CNN features
1063.2DQ-Mask2Former
ResNet-50
Learning Dynamic Query Combinations for Transformer-based Object Detection and Segmentation
ICML 2023 · CityScapes segmentation table
CNN features
17 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

Cityscapes · FID

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

13 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

Cityscapes · AP (val)

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

#AP ↑MethodPaperTopic
146.9PolyTransform + BPR
PolyTransform
Look Closer To Segment Better: Boundary Patch Refinement for Instance Segmentation
CVPR 2021 · Table 10
CNN features
245.6OneFormer
Swin-L
OneFormer: One Transformer To Rule Universal Image Segmentation
CVPR 2023 · Table 2
CNN features
344.6PolyTransform
UPSNet
PolyTransform: Deep Polygon Transformer for Instance Segmentation
CVPR 2020 · Table: Comparison with naive r
CNN features
444kMaX-DeepLab
ConvNeXt-L (ImageNet-22K)
k-Means Mask Transformer
ECCV 2022 · Table 2
CNN features
538.93OneFormer + synthesis
ConvNext-L
A Generative Approach at the Instance-Level for Image Segmentation Under Limited Training Data Conditions (Student Abstract)
AAAI 2025 · Table 2
Open-set and few-shot detection
638.3BoundaryFormer
COCO initialization
Instance Segmentation With Mask-Supervised Polygonal Boundary Transformers
CVPR 2022 · Table 3
CNN features
738.2PatchDCT
R50-FPN
PatchDCT: Patch Refinement for High Quality Instance Segmentation
ICLR 2023 · Cityscapes results table
CNN features
837.7Mask R-CNN + SharpContour
Mask R-CNN
SharpContour: A Contour-Based Boundary Refinement Approach for Efficient and Accurate Instance Segmentation
CVPR 2022 · Table 2
CNN features
937.6RefineMask
R50-FPN
RefineMask: Towards High-Quality Instance Segmentation With Fine-Grained Features
CVPR 2021 · Table 9
CNN features
1037.4Deep snake
CenterNet
Deep Snake for Real-Time Instance Segmentation
CVPR 2020 · Cityscapes results table
CNN features
11 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

Cityscapes · AP

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

#AP ↑MethodPaperTopic
172.9LEGOLEGO: Learning Edge With Geometry All at Once by Watching Videos
CVPR 2018 · Edge experiments table
Depth and stereo
249.5Seg-VARSeg-VAR:Image Segmentation with Visual Autoregressive Modeling
NeurIPS 2025 · Table: Results on Image Segmen
Autoregressive and latent generation
342.7PolyTransform + SegFix + BPR
PolyTransform + SegFix
Look Closer To Segment Better: Boundary Patch Refinement for Instance Segmentation
CVPR 2021 · Table 10
CNN features
440.8Sparse R-CNN + Spearman Correlation Loss
Sparse R-CNN
Correlation Loss: Enforcing Correlation between Classification and Localization
AAAI 2023 · Cityscapes results table
CNN features
540.1PolyTransform
UPSNet
PolyTransform: Deep Polygon Transformer for Instance Segmentation
CVPR 2020 · Table: Instance segmentation o
CNN features
636.4PANet [COCO]
ResNet-50
Path Aggregation Network for Instance Segmentation
CVPR 2018 · Cityscapes results table
CNN features
732.8SP
QueryInst
Boosting Low-Data Instance Segmentation by Unsupervised Pre-Training With Saliency Prompt
CVPR 2023 · Table: Instance segmentation f
Open-set and few-shot detection
832.7SSAPSSAP: Single-Shot Instance Segmentation With Affinity Pyramid
ICCV 2019 · Table 9
CNN features
931.7Deep snake
CenterNet
Deep Snake for Real-Time Instance Segmentation
CVPR 2020 · Cityscapes results table
CNN features
1029.5DELA-CondDETRTowards Data-Efficient Detection Transformers
ECCV 2022 · Table 2
CNN features
10 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

Cityscapes · AP50

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

#AP50 ↑MethodPaperTopic
165.9PolyTransform
UPSNet
PolyTransform: Deep Polygon Transformer for Instance Segmentation
CVPR 2020 · Table: Instance segmentation o
CNN features
258.4Deep snake
CenterNet
Deep Snake for Real-Time Instance Segmentation
CVPR 2020 · Cityscapes results table
CNN features
353.6KGD
Detic
Domain Adaptation for Large-Vocabulary Object Detectors
NeurIPS 2024 · Benchmarking table
Open-set and few-shot detection
451.8SSAPSSAP: Single-Shot Instance Segmentation With Affinity Pyramid
ICCV 2019 · Table 9
CNN features
545.6Ours (affinity + graph merge)Affinity Derivation and Graph Merge for Instance Segmentation
ECCV 2018 · Table 1
CNN features
644.9SGN
PSP + VGG16 breakpoint net
SGN: Sequential Grouping Networks for Instance Segmentation
ICCV 2017 · Table 1
CNN features
739Multi-Task Learning
ResNet101
Multi-Task Learning Using Uncertainty to Weigh Losses for Scene Geometry and Semantics
CVPR 2018 · Table 2
Recognition and long-tail
836.7BAIS - full
VGG16
Boundary-Aware Instance Segmentation
CVPR 2017 · Table 3
CNN features
935.3Deep Watershed TransformDeep Watershed Transform for Instance Segmentation
CVPR 2017 · Table 1
CNN features
1027.9InstanceCutInstanceCut: From Edges to Instances With MultiCut
CVPR 2017 · Table 1
CNN features
1 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

Cityscapes · FPS

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

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