List your company
ResearchFormerly WizWand

Referring Expression Comprehension on RefCOCO v1 (val)

92.83Top-1 AccuracyBARE-L
48.245259.820171.39582.9699Apr 7, 2024Jul 27, 2024Nov 15, 2024Mar 6, 2025Jun 25, 2025Oct 14, 2025Feb 2, 2026
Updated 8mo ago

Evaluation results

MethodLinks
BARE-LVisual Encoder=BEiT-L [9], Fine-tuning protocol=Fine-tuning, Pre-training type=pretrained open-set detection model, Training dataset scope=mixed datasets2026.0192.83
ObjEmbedParameters=4B2026.0292.5
BARE-LVisual Encoder=BEiT-L [9], Fine-tuning protocol=Fine-tuning, Pre-training type=pretrained vision-language model, Training dataset scope=single dataset2026.0192.37
ObjEmbedParameters=2B2026.0291.7
CLARE-LSize=Large2026.0291.4
VLM-FO1Parameters=3B2026.0291.1
ChatRexParameters=7B2026.0291
HyperLearnerPre-training Data=O365, GoldG, RefC, Fine-tuning=true2024.0490.74
Qwen3-VLParameters=4B2026.0290.7
SimVG-LVenue=NeurIPS’24, Visual Encoder=BEiT-L [9], Fine-tuning protocol=Fine-tuning, Pre-training type=pretrained vision-language model, Training dataset scope=single dataset2026.0190.61
Grounding-DINO-LSize=Large2026.0290.6
InternVL2.5Parameters=8B2026.0290.3
InternVL3.5Parameters=38B2026.0290.3
BARE-BVisual Encoder=BEiT-B [9], Fine-tuning protocol=Fine-tuning, Pre-training type=pretrained open-set detection model, Training dataset scope=mixed datasets2026.0190.28
VLM-R1Parameters=3B2026.0290.1
OFA-LVenue=ICML’22, Visual Encoder=OFA-L [66], Fine-tuning protocol=Fine-tuning, Pre-training type=pretrained open-set detection model, Training dataset scope=mixed datasets2026.0190.05
Qwen2.5-VLParameters=7B2026.0290
Grounding-DINO-TPre-training Data=O365, GoldG, RefC, Fine-tuning=true2024.0489.19
Grounding-DINOVenue=ECCV’24, Visual Encoder=Swin-T [12], Fine-tuning protocol=Fine-tuning, Pre-training type=pretrained open-set detection model, Training dataset scope=mixed datasets2026.0189.19
Qwen2.5-VLParameters=3B2026.0289.1
PLVLVenue=arXiv’25, Visual Encoder=ViT-B [29], Fine-tuning protocol=Fine-tuning, Pre-training type=pretrained vision-language model, Training dataset scope=single dataset2026.0189.02
OctopusParameters=7B2026.0289
BARE-BVisual Encoder=BEiT-B [9], Fine-tuning protocol=Fine-tuning, Pre-training type=pretrained vision-language model, Training dataset scope=single dataset2026.0188.69
DQ-DETRPre-training Data=GoldG, RefC, Fine-tuning=true2024.0488.63
DQ-DETRVenue=AAAI’23, Visual Encoder=RN101 [11], Fine-tuning protocol=Fine-tuning, Pre-training type=pretrained open-set detection model, Training dataset scope=mixed datasets2026.0188.63
OFA-BVenue=ICML’22, Visual Encoder=OFA-B [66], Fine-tuning protocol=Fine-tuning, Pre-training type=pretrained open-set detection model, Training dataset scope=mixed datasets2026.0188.48
Qwen3-VLParameters=2B2026.0288.2
HiVG-LVenue=ACM MM’24, Visual Encoder=CLIP-L [29], Fine-tuning protocol=Fine-tuning, Pre-training type=pretrained vision-language model, Training dataset scope=single dataset2026.0188.14
SimVG-BVenue=NeurIPS’24, Visual Encoder=BEiT-B [9], Fine-tuning protocol=Fine-tuning, Pre-training type=pretrained vision-language model, Training dataset scope=single dataset2026.0187.63
HiVG-BVenue=ACM MM’24, Visual Encoder=CLIP-B [29], Fine-tuning protocol=Fine-tuning, Pre-training type=pretrained vision-language model, Training dataset scope=single dataset2026.0187.32
DUETVenue=TCSVT’24, Visual Encoder=Swin-S [12], Fine-tuning protocol=Fine-tuning, Pre-training type=pretrained vision-language model, Training dataset scope=single dataset2026.0187.27
SegVGVenue=ECCV’24, Visual Encoder=Swin-S [12], Fine-tuning protocol=Fine-tuning, Pre-training type=pretrained close-set detection and language model, Training dataset scope=single dataset2026.0186.84
MFSDVenue=TCSVT’24, Visual Encoder=RN101 [11], Fine-tuning protocol=Fine-tuning, Pre-training type=pretrained close-set detection and language model, Training dataset scope=single dataset2026.0186.82
MDETRPre-training Data=GoldG, RefC, Fine-tuning=true2024.0486.75
Rex-OmniParameters=3B2026.0286.6
UniTABVenue=ECCV’22, Visual Encoder=RN101 [11], Fine-tuning protocol=Fine-tuning, Pre-training type=pretrained open-set detection model, Training dataset scope=mixed datasets2026.0186.32
VG-LAWVenue=CVPR’23, Visual Encoder=ViT-B [29], Fine-tuning protocol=Fine-tuning, Pre-training type=pretrained close-set detection and language model, Training dataset scope=single dataset2026.0186.06
RefTRPre-training Data=VG, Fine-tuning=true2024.0485.65
VLTVGVenue=CVPR’22, Visual Encoder=RN101 [11], Fine-tuning protocol=Fine-tuning, Pre-training type=pretrained close-set detection and language model, Training dataset scope=single dataset2026.0184.77
CLIP-VGVenue=TMM’23, Visual Encoder=CLIP-B [29], Fine-tuning protocol=Fine-tuning, Pre-training type=pretrained vision-language model, Training dataset scope=single dataset2026.0184.29
QRNetVenue=CVPR’22, Visual Encoder=Swin-S [12], Fine-tuning protocol=Fine-tuning, Pre-training type=pretrained close-set detection and language model, Training dataset scope=single dataset2026.0184.01
YOROVenue=ECCV’22, Visual Encoder=ViLT [43], Fine-tuning protocol=Fine-tuning, Pre-training type=pretrained open-set detection model, Training dataset scope=mixed datasets2026.0182.9
HyperLearnerPre-training Data=O365, GoldG, RefC, Fine-tuning=false2024.0477.89
MAttNetPre-training Data=RefC, Fine-tuning=true2024.0476.65
Grounding-DINO-TPre-training Data=O365, GoldG, RefC, Fine-tuning=false2024.0473.98
HyperLearnerPre-training Data=O365, GoldG, Fine-tuning=false2024.0450.66
GLIP-T(C)Pre-training Data=O365, GoldG, Cap4M, Fine-tuning=false2024.0450.42
Grounding-DINO-TPre-training Data=O365, GoldG, Fine-tuning=false2024.0450.41
GLIP-T(B)Pre-training Data=O365, GoldG, Fine-tuning=false2024.0449.96