AI Paper Atlas

Research map / Benchmarks / COCO

COCO leaderboard

782 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on COCO. Best AP (test-dev): 81.1 (ViTPose-G ensemble, TPAMI 2023).

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.

COCO · AP (test-dev)

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

#AP ↑MethodPaperTopic
181.1ViTPose-G ensembleViTPose++: Vision Transformer for Generic Body Pose Estimation
TPAMI 2023 · Table 12
Human pose estimation
280.9ViTPose
ViTAE-G
ViTPose: Simple Vision Transformer Baselines for Human Pose Estimation
NeurIPS 2022 · Table: Comparison with SOTA me
Attention and transformers
380.6SCAI
HRNet HR48
Self-Correctable and Adaptable Inference for Generalizable Human Pose Estimation
CVPR 2023 · Table: Comparison with the sta
Recognition and long-tail
479.2SCIO (HR48)
HRNet-W48
Self-Constrained Inference Optimization on Structural Groups for Human Pose Estimation
ECCV 2022 · Table 1
Human pose estimation
579.2RSN* ensemble
4xRSN-50
Learning Delicate Local Representations for Multi-Person Pose Estimation
ECCV 2020 · Table 5
Human pose estimation
678.3Poseur (6 Dec.)
HRFormer-B
Poseur: Direct Human Pose Regression with Transformers
ECCV 2022 · Table 8
Human pose estimation
777.5SSPCM (+DARK)
HRNetW48, 384 × 288
Semi-Supervised 2D Human Pose Estimation Driven by Position Inconsistency Pseudo Label Correction Module
CVPR 2023 · Table: comparison on COCO TEST
Semi-supervised learning
877.2Ours (+DARK)
HRNetW48
An Empirical Study of the Collapsing Problem in Semi-Supervised 2D Human Pose Estimation
ICCV 2021 · Table 4
Human pose estimation
976.5HRNet-W48-384×288 + UDP
HRNet-W48
The Devil Is in the Details: Delving Into Unbiased Data Processing for Human Pose Estimation
CVPR 2020 · Table: COCO test-dev results
Human pose estimation
1076.4DARK + ReLoss
HRNet-W48
Relational Surrogate Loss Learning
ICLR 2022 · Table Results of human pose es
Prediction and decision losses
141 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

COCO · AP (val)

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

#AP ↑MethodPaperTopic
177.2HRNet-W48-256×192 + UDP
HRNet-W48
The Devil Is in the Details: Delving Into Unbiased Data Processing for Human Pose Estimation
CVPR 2020 · Table: COCO val results
Human pose estimation
276.8HRNet-W48 (384x288) + PoseTrans
HRNet-W48 (384x288)
PoseTrans: A Simple yet Effective Pose Transformation Augmentation for Human Pose Estimation
ECCV 2022 · Table 1
Human pose estimation
375.8+ BCIR
HRNet-W32
Bias-Compensated Integral Regression for Human Pose Estimation
TPAMI 2023 · Table (COCO val)
Human pose estimation
475.8Bias-compensated joint method (Ours)
HRNet-W32 256x192
Removing the Bias of Integral Pose Regression
ICCV 2021 · Table 2
Human pose estimation
574.8HRNet + ReLoss
HRNet-W32
Relational Surrogate Loss Learning
ICLR 2022 · Table Results of human pose es
Prediction and decision losses
674.4DistilPose-L
HRNet-W48-stage3
DistilPose: Tokenized Pose Regression With Heatmap Distillation
CVPR 2023 · Table 3
Knowledge distillation
774.14-stage IMHN plus with validation data
4-stage IMHN plus
Simple Pose: Rethinking and Improving a Bottom-up Approach for Multi-Person Pose Estimation
AAAI 2020 · Table 1
Human pose estimation
873.3DETRPose-X
HGNetv2-B5
DETRPose: Real-Time End-to-End Multi-Person Pose Estimation via Modified Transformer Decoder and Novel Denoising Keypoints
ECCV 2026 · Table 3
Human pose estimation
972.1MA-refine (PoseFix)
CPN
PoseFix: Model-Agnostic General Human Pose Refinement Network
CVPR 2019 · Table (validation refinement c
Human pose estimation
1070.8SIMPLE's implementation
SIMPLE-W32
SIMPLE: SIngle-network with Mimicking and Point Learning for Bottom-up Human Pose Estimation
AAAI 2021 · Ablation study of network mimi
Human pose estimation
96 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

COCO · mAP

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

#mAP ↑MethodPaperTopic
197.75CMLL
ViT-B/32
Towards Reliable Multi-Label Classification via Conditional Dependency Modeling
ECCV 2026 · Table 2
Recognition and long-tail
292.8MAT-AgentMAT-Agent: Adaptive Multi-Agent Training Optimization
NeurIPS 2025 · Table 1
Human-AI and goal-conditioned
390.3CCD-Swin
Swin-L(22k)
Contextual Debiasing for Visual Recognition With Causal Mechanisms
CVPR 2022 · Table 1
Recognition and long-tail
488.3PatchCT
ViT-B/16 + BERT
PatchCT: Aligning Patch Set and Label Set with Conditional Transport for Multi-Label Image Classification
ICCV 2023 · Table 1
Recognition and long-tail
587MGTN(FINAL)
Dual ResNeXt-50 32x4d
Modular Graph Transformer Networks for Multi-Label Image Classification
AAAI 2021 · Table 1
Graph self-supervised learning
684.9Ours
ResNet-101
In Pursuit of Causal Label Correlations for Multi-label Image Recognition
NeurIPS 2024 · Table 8
Recognition and long-tail
783.8MS-CMA
ResNet-101
Cross-Modality Attention with Semantic Graph Embedding for Multi-Label Classification
AAAI 2020 · Table 1
Recognition and long-tail
882.2Semantic softmax + KD
TResNet-M
ImageNet-21K Pretraining for the Masses
NeurIPS 2021 · Table, Comparing downstream re
Attention and transformers
981.9VisionLLM v2
Vicuna-7B; Swin-T
VisionLLM v2: An End-to-End Generalist Multimodal Large Language Model for Hundreds of Vision-Language Tasks
NeurIPS 2024 · Region Recognition table
Multimodal LLMs
1081LaSO unionLaSO: Label-Set Operations Networks for Multi-Label Few-Shot Learning
CVPR 2019 · Table 3
Zero-/few-shot and meta-learning
64 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

COCO · FID

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

#FID ↓MethodPaperTopic
14.87DiCo-CrossAttnDiCo: Revitalizing ConvNets for Scalable and Efficient Diffusion Modeling
NeurIPS 2025 · Text-to-image comparison table
Autoregressive and latent generation
25.6UViT-S/2 + LS
UViT-S/2
ScaleLong: Towards More Stable Training of Diffusion Model via Scaling Network Long Skip Connection
NeurIPS 2023 · Table 1
RNNs and normalization
35.85GALIP
CLIP ViT-B/32
GALIP: Generative Adversarial CLIPs for Text-to-Image Synthesis
CVPR 2023 · Table 1
GANs
46.29TLDM (NFE=250)
LDM
Truncated Diffusion Probabilistic Models and Diffusion-based Adversarial Auto-Encoders
ICLR 2023 · Text-to-image results table
Fast diffusion sampling
56.78Transfusion 7B, 2T tokensTransfusion: Predict the Next Token and Diffuse Images with One Multi-Modal Model
ICLR 2025 · Table: comparison with image g
Human and embodied modeling
66.82ENAT-B (8 steps)ENAT: Rethinking Spatial-temporal Interactions in Token-based Image Synthesis
NeurIPS 2024 · Table: Results on MS-COCO
Autoregressive and latent generation
77.24MANGOMANGO: Unleashing Image Generation Capability of Unified Multimodal Models
ECCV 2026 · Table 4
Multimodal LLMs
88.109Reject decoding coco, Nref=10
RQ-Transformer coco
Reject Decoding via Language-Vision Models for Text-to-Image Synthesis
AAAI 2023 · Table 1
Human and embodied modeling
98.38Harmon-H + Harmon-B draft
Harmon-1.5B
Continuous Speculative Decoding for Autoregressive Image Generation
ECCV 2026 · Table 6
Fast diffusion sampling
108.5Setokim
LLaMA-2-7B
Towards Semantic Equivalence of Tokenization in Multimodal LLM
ICLR 2025 · Table: generation and editing
Multimodal LLMs
48 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

COCO · AP

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

41 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

COCO · CIDEr

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

#CIDEr ↑MethodPaperTopic
1204.2OursSay As You Wish: Fine-Grained Control of Image Caption Generation With Abstract Scene Graphs
CVPR 2020 · Table: Comparison with careful
Knowledge and temporal graphs
2149.3AKGMA
Qwen2.5-7B
Knowledge-Enhanced Image Captioning with Adaptive Graph-based Multimodal Alignment and LLM
AAAI 2026 · Table 1
Human and embodied modeling
3143.3BEiT-3 (Large)-OTMV
BEiT-3 (Large)
Once-Tuning-Multiple-Variants: Tuning Once and Expanded as Multiple Vision-Language Model Variants
CVPR 2025 · Table 1
Multimodal LLMs
4142.8Chain-of-Sight, 80 pre-training / 1296 fine-tuning tokens
Vicuna
Accelerating Pre-training of Multimodal LLMs via Chain-of-Sight
NeurIPS 2024 · Table 1
Multimodal LLMs
5142.6CW-DPO (Ours)
Qwen2.5-VL-72B
Dynamics-Aware Preference Optimization for Vision-Language Models
CVPR 2026 · Table 2
Multimodal LLMs
6140.1EVCAP (Vicuna-13B)
Vicuna-13B
EVCap: Retrieval-Augmented Image Captioning with External Visual-Name Memory for Open-World Comprehension
CVPR 2024 · Table 1
CLIP and zero-shot
7137.4SGDiffSGDiff: Scene Graph Guided Diffusion Model for Image Collaborative SegCaptioning
AAAI 2025 · Table 1
Open-set and few-shot detection
8133.5Ours + MM-SeRMM-SeR: Multimodal Self-Refinement for Lightweight Image Captioning
CVPR 2026 · Table 3
Multimodal LLMs
9131.7Uncertainty order
UAIC
Uncertainty-Aware Image Captioning
AAAI 2023 · Table 3
Human and embodied modeling
10129.4PNAIC (K = 2)
Transformer
Partially Non-Autoregressive Image Captioning
AAAI 2021 · Table 2
Human and embodied modeling
36 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

COCO · AP (minival)

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

#AP ↑MethodPaperTopic
155.8DensePose + keypoints (cascade)
ResNet-50
DensePose: Dense Human Pose Estimation in the Wild
CVPR 2018 · Table
Human pose estimation
251.3AdaMixer
Swin-S, 300 queries, 3×
AdaMixer: A Fast-Converging Query-Based Object Detector
CVPR 2022 · text
CNN features
349.1GAIA-TSAS
GAIA-TSAS
GAIA: A Transfer Learning System of Object Detection That Fits Your Needs
CVPR 2021 · Table: Results of models with
Open-set and few-shot detection
448.3AIRS
GFocal, ResNet-50
Adaptive Important Region Selection with Reinforced Hierarchical Search for Dense Object Detection
NeurIPS 2024 · Table 1
CNN features
546.7D-DETR+Ours
Deformable DETR
Progressive End-to-End Object Detection in Crowded Scenes
CVPR 2022 · Table 4
CNN features
644.7Universal-RCNN
FPN, ResNet-101
Universal-RCNN: Universal Object Detector via Transferable Graph R-CNN
AAAI 2020 · Table 1
Open-set and few-shot detection
743.9CrossKD+PKD
GFL-R50
CrossKD: Cross-Head Knowledge Distillation for Object Detection
CVPR 2024 · Table 5
Knowledge distillation
843.9TSD + progressive constraint
ResNet-152
Revisiting the Sibling Head in Object Detector
CVPR 2020 · Table: variant backbones
CNN features
942.8FreeAnchor + SEPC
FreeAnchor, ResNet-50
Scale-Equalizing Pyramid Convolution for Object Detection
CVPR 2020 · Table: Comparison of detection
CNN features
1042.7D2Det
ResNet50 with FPN
D2Det: Towards High Quality Object Detection and Instance Segmentation
CVPR 2020 · Table 2
CNN features
34 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

COCO · mAP (val)

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

30 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

COCO · FID (val)

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

#FID ↓MethodPaperTopic
15.69DPM-Solver++ (3Fast) + MASF
U-ViT
Boosting Diffusion Models with Moving Average Sampling in Frequency Domain
CVPR 2024 · Table 2
Fast diffusion sampling
28.55OursSceneComposer: Any-Level Semantic Image Synthesis
CVPR 2023 · Table 2
Text-to-image diffusion
39.71UniDiffuserOne Transformer Fits All Distributions in Multi-Modal Diffusion at Scale
ICML 2023 · Table 1
Text-to-image diffusion
411.66Emu
LLaMA-13B; Stable Diffusion v1.5 decoder initialization
Emu: Generative Pretraining in Multimodality
ICLR 2024 · Zero-shot text-to-image result
CLIP and zero-shot
512.1Smooth Diffusion
Stable Diffusion v1.5
Smooth Diffusion: Crafting Smooth Latent Spaces in Diffusion Models
CVPR 2024 · Table 1
Editing and style transfer
614.557Ours
Stable Diffusion 1.5
Devil is in the Detail: Towards Injecting Fine Details of Image Prompt in Image Generation via Conflict-free Guidance and Stratified Attention
CVPR 2025 · Table 1
Text-to-image diffusion
715.64FreeControl (One Step Extraction)
FLUX.1-dev
FreeControl: Efficient, Training-Free Structural Control via One-Step Attention Extraction
NeurIPS 2025 · Comparison with controlled-gen
Fast diffusion sampling
817.45RaSS (α = 1.5)
Stable Diffusion
RaSS: Improving Denoising Diffusion Samplers with Reinforced Active Sampling Scheduler
CVPR 2025 · Table 3
Fast diffusion sampling
919.15TokenCompose
Stable Diffusion 2.1
TokenCompose: Text-to-Image Diffusion with Token-level Supervision
CVPR 2024 · Table 3
Text-to-image diffusion
1019.28DetDiffusion origin
Stable Diffusion v1.5
DetDiffusion: Synergizing Generative and Perceptive Models for Enhanced Data Generation and Perception
CVPR 2024 · Table 1
Fast diffusion sampling
19 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

COCO · mask AP (test-dev)

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

#mask AP ↑MethodPaperTopic
151.8UNINEXT (ViT-H)
ViT-H
Universal Instance Perception As Object Discovery and Retrieval
CVPR 2023 · Table 2
Open-set and few-shot detection
249.2SOIT
Swin-L
SOIT: Segmenting Objects with Instance-Aware Transformers
AAAI 2022 · Table 1
CNN features
349.1Self-training Copy-Paste
Cascade Eff-B7 NAS-FPN
Simple Copy-Paste Is a Strong Data Augmentation Method for Instance Segmentation
CVPR 2021 · Table: Comparison with the sta
Open-set and few-shot detection
448.7SOLQ + Ours
SOLQ Swin-L
Learning Equivariant Segmentation with Instance-Unique Querying
NeurIPS 2022 · Table 8
CNN features
548.5DetectoRS
ResNeXt-101-64x4d
DetectoRS: Detecting Objects With Recursive Feature Pyramid and Switchable Atrous Convolution
CVPR 2021 · Table: instance segmentation c
CNN features
646.7SOLQ
Swin-L
SOLQ: Segmenting Objects by Learning Queries
NeurIPS 2021 · Table 1
CNN features
745.5DSC (multi-scale training)
ResNeXt-101-64x4d DCN FPN
Deeply Shape-Guided Cascade for Instance Segmentation
CVPR 2021 · Table: Comparison with state-o
CNN features
845EXITS
Swin-Small + Mask2Former
Extreme Point Supervised Instance Segmentation
CVPR 2024 · Table 1
Weakly supervised detection and segmentation
944.6Transformer-based BCNet
R50-FPN
Occlusion-Aware Instance Segmentation Via BiLayer Network Architectures
TPAMI 2023 · Table 11
CNN features
1042.8SOLOv2+
R-dcn101-FPN
SOLO: A Simple Framework for Instance Segmentation
TPAMI 2021 · Table 1
CNN features
19 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

COCO · mIoU

26 papers, each with its best reported number for its own method; higher 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

COCO · mIoU (val)

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

#mIoU ↑MethodPaperTopic
175SAM-PARSER
SAM ViT-Base
SAM-PARSER: Fine-Tuning SAM Efficiently by Parameter Space Reconstruction
AAAI 2024 · Main results table
Pruning and efficient tuning
270.9Fixed ratio 9:7
HRNetV2-W48
Weakly-Supervised Instance Segmentation via Class-Agnostic Learning With Salient Images
CVPR 2021 · Table: Results using different
Weakly supervised detection and segmentation
367.3kMaX-DeepLab trained on COCONut-L
ConvNeXt-L
COCONut: Modernizing COCO Segmentation
CVPR 2024 · Table 6
Open-set and few-shot detection
456.8DHR (Mask2Former)
Swin-L
DHR: Dual Features-Driven Hierarchical Rebalancing in Inter- and Intra-Class Regions for Weakly-Supervised Semantic Segmentation
ECCV 2024 · Table 2
Weakly supervised detection and segmentation
554.5LPSNet with PPM
ResNet-50
LPSNet: A Lightweight Solution for Fast Panoptic Segmentation
CVPR 2021 · Table 5
CNN features
650.6SSR (Ours)SSR: Semantic and Spatial Rectification for CLIP-based Weakly Supervised Segmentation
AAAI 2026 · Table 1
Weakly supervised detection and segmentation
750.3ExCEL
CLIP ViT-B
Exploring CLIP's Dense Knowledge for Weakly Supervised Semantic Segmentation
CVPR 2025 · Table 1
CLIP and zero-shot
847.4MoRe
ViT-B
MoRe: Class Patch Attention Needs Regularization for Weakly Supervised Semantic Segmentation
AAAI 2025 · Table 2
Weakly supervised detection and segmentation
946.7SeCo
ViT-B/16
Separate and Conquer: Decoupling Co-occurrence via Decomposition and Representation for Weakly Supervised Semantic Segmentation
CVPR 2024 · Table 1(b)
Weakly supervised detection and segmentation
1046.6MuP-VSS
RN38
Multi-Label Prototype Visual Spatial Search for Weakly Supervised Semantic Segmentation
CVPR 2025 · Table 2
Weakly supervised detection and segmentation
14 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

24 less reported leaderboards for COCO (other metrics or splits) are in the CSV and the API.

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