AI Paper Atlas

Research map / Benchmarks / RefCOCO testA

RefCOCO testA leaderboard

91 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on RefCOCO testA. Best accuracy: 96.3 (RTAF (Ours), ICML 2026).

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.

RefCOCO testA · accuracy

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

#accuracy ↑MethodPaperTopic
196.3RTAF (Ours)
Qwen2.5-VL-7B
Reason with Thumbnails, Answer with Focus: An Efficient and Effective Paradigm for Multimodal Grounded Visual Reasoning
ICML 2026 · Table 2
Multimodal LLMs
296IVC-Prune (52% tokens)
DeepSeek-VL2-Small-16B
IVC-Prune: Revealing the Implicit Visual Coordinates in LVLMs for Vision Token Pruning
ICLR 2026 · Table 1
Multimodal LLMs
395.3Florence-2-LFlorence-2: Advancing a Unified Representation for a Variety of Vision Tasks
CVPR 2024 · Table 5
CLIP and zero-shot
494.57CogCoM
CogVLM
CogCoM: A Visual Language Model with Chain-of-Manipulations Reasoning
ICLR 2025 · Table: Results on VG benchmark
Multimodal LLMs
594.04SSRVG-B*
Swin-B
Look Around Before Locating: Considering Content and Structure Information for Visual Grounding
AAAI 2025 · Table 1
Human and embodied modeling
693.23AerialVG (Ours)
Swin-L
AerialVG: A Challenging Benchmark for Aerial Visual Grounding by Exploring Positional Relations
ICCV 2025 · Table 2
Human and embodied modeling
792.91CogVLM-GroundingCogVLM: Visual Expert for Pretrained Language Models
NeurIPS 2024 · Grounding table
Human and embodied modeling
892.9IAA-14
Llama3-8B
IAA: Inner-Adaptor Architecture Empowers Frozen Large Language Model with Multimodal Capabilities
AAAI 2025 · Table: Comparisons on visual g
Multimodal LLMs
992.63Octopus
Vicuna-7B-v1.5
Octopus: A Multi-modal LLM with Parallel Recognition and Sequential Understanding
NeurIPS 2024 · Table 1
Multimodal LLMs
1092.12Elysium (7B)
7B MLLM, 336px, 108 tokens
Elysium: Exploring Object-level Perception in Videos through Semantic Integration Using MLLMs
ECCV 2024 · Table 2
Video-language understanding
29 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

RefCOCO testA · mIoU

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

#mIoU ↑MethodPaperTopic
186.3FOCUS
Qwen2.5-3B
FOCUS: Unified Vision-Language Modeling for Interactive Editing Driven by Referential Segmentation
NeurIPS 2025 · Referring segmentation table
Multimodal LLMs
283.7GenMask
WAN-2.1 DiT 1.3B + Qwen2.5-VL-7B
GenMask: Adapting DiT for Segmentation via Direct Mask Generation
CVPR 2026 · Table 1
Autoregressive and latent generation
381.6UniRES
CLIP
Unveiling Parts Beyond Objects: Towards Finer-Granularity Referring Expression Segmentation
CVPR 2024 · Table 4
Open-set and few-shot detection
479.7Ours-LWIMFRIS: WIndow Mamba Fusion and Parameter Efficient Tuning for Referring Image Segmentation
ICLR 2026 · Table 2
CNN features
578.12TCRT
Swin-B
Task-aware Cross-modal Feature Refinement Transformer with Large Language Models for Visual Grounding
CVPR 2025 · Table 2
Human and embodied modeling
677.92RefFormerReferencing Where to Focus: Improving Visual Grounding with Referential Query
NeurIPS 2024 · Table 2
Human and embodied modeling
776.89LAVT
Swin-B + BERT
Language-Aware Vision Transformer for Referring Segmentation
TPAMI 2024 · Table 7
Cross-modal learning
863.81DViN
YOLOv5-Seg
DViN: Dynamic Visual Routing Network for Weakly Supervised Referring Expression Comprehension
CVPR 2025 · Table 6
Human and embodied modeling
961.18WeakMCN†WeakMCN: Multi-task Collaborative Network for Weakly Supervised Referring Expression Comprehension and Segmentation
CVPR 2025 · Table 1
Weakly supervised detection and segmentation
1053.37Ours
SAM, CLIP ViT-B/16
Hybrid Global-Local Representation with Augmented Spatial Guidance for Zero-Shot Referring Image Segmentation
CVPR 2025 · Table 1
CLIP and zero-shot
6 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

RefCOCO testA · cIoU

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

#cIoU ↑MethodPaperTopic
186.6ALToLLM-8B (AL)
InternVL-2.5-8B
ALTo: Adaptive-Length Tokenizer for Autoregressive Mask Generation
NeurIPS 2025 · Table: Performance comparison
Multimodal LLMs
286.6InstructSeg
Mipha-3B
InstructSeg: Unifying Instructed Visual Segmentation with Multi-modal Large Language Models
ICCV 2025 · Table 2
Video-language understanding
385.3CoPRS-7B
Qwen2.5-VL-7B
CoPRS: Learning Positional Prior from Chain-of-Thought for Reasoning Segmentation
ICLR 2026 · Table 1
Multimodal LLMs
482DenseMLLM
DenseMLLM-4B
DenseMLLM: Standard Multimodal LLMs for Dense Prediction
ICML 2026 · Table 1
Multimodal LLMs
581.44SegAgent-LLaVA+SAM
LLaVA-v1.5-7B
SegAgent: Exploring Pixel Understanding Capabilities in MLLMs by Imitating Human Annotator Trajectories
CVPR 2025 · RES comparison table
Multimodal LLMs
680.8SAM-Veteran
Qwen2.5-VL-7B
SAM-Veteran: An MLLM-Based Human-like SAM Agent for Reasoning Segmentation
ICLR 2026 · Table 3
Multimodal LLMs
780.3DGSeg
Qwen2.5-VL-7B
DGSeg: Dynamic Gating of Semantic-Spatial Guided Predictions for Reasoning Segmentation
ECCV 2026 · Table 2
Multimodal LLMs
880.2READ
LLaVA1.5-7B
Reasoning to Attend: Try to Understand How <SEG> Token Works
CVPR 2025 · RefCOCO(+/g) results table
Multimodal LLMs
979.5Vitron
Vicuna 7B v1.5
Vitron: A Unified Pixel-level Vision LLM for Understanding, Generating, Segmenting, Editing
NeurIPS 2024 · Table: Results of referring im
Multimodal LLMs
1079.3PixelThink-7B
Qwen2.5-VL-7B
Don't Overthink with Pixels: Efficient Reasoning for Segmentation
ICML 2026 · Table 2b
Multimodal LLMs
4 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

RefCOCO testA · Acc@0.5

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

3 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

RefCOCO testA · oIoU

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

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.