Referring Expression Counting on REC 8K (val)
4.23MAECAD-GD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CAD-GDBackbone=Swin-B2026.06 | 4.23 | 13.14 | |
| Qwen2.5VL-7B-Instruct + REC-RLTraining Dataset=Subset-2K2026.05 | 5.05 | 15.87 | |
| DCountBackbone=Swin-T2026.06 | 5.24 | 15.47 | |
| MambaCount2026.06 | 5.28 | 16.26 | |
| Qwen2.5VL-7B-Instruct + FGRPRTraining Dataset=Subset-2K2026.05 | 5.29 | 16.2 | |
| CAD-GDBackbone=Swin-T2026.06 | 5.43 | 15.01 | |
| Qwen2.5VL-3B-Instruct + REC-RLTraining Dataset=Subset-2K2026.05 | 5.48 | 16.27 | |
| Qwen2.5VL-3B-Instruct + FGRPRTraining Dataset=Subset-2K2026.05 | 5.65 | 17.41 | |
| Ground-D RECTraining Dataset=all data2026.05 | 6.8 | 18.13 | |
| GroundingREC2026.05 | 6.8 | 18.13 | |
| GroundingRECBackbone=Swin-T2026.06 | 6.8 | 18.13 | |
| Qwen2.5VL-7B-Instruct + SFTTraining Dataset=all data2026.05 | 8.27 | 20.64 | |
| Qwen2.5VL-3B-Instruct + SFTTraining Dataset=all data2026.05 | 8.63 | 22.66 | |
| Qwen2.5VL-7B-InstructTraining Dataset=None2026.05 | 9.02 | 22.6 | |
| Ground-D Swin-TTraining Dataset=all data2026.05 | 9.03 | 21.98 | |
| GDino2026.05 | 9.03 | 21.98 | |
| GroundingDINOBackbone=Swin-T2026.06 | 9.03 | 21.98 | |
| Qwen2.5VL-3B-InstructTraining Dataset=None2026.05 | 9.24 | 22.54 | |
| VisHarnesstraining=GRPO2026.05 | 9.31 | 23.16 | |
| CountGD2026.05 | 9.51 | 22.91 | |
| VisHarness-SFTtraining=SFT2026.05 | 9.89 | 28.16 | |
| Ground-D Swin-TTraining Dataset=None2026.05 | 11.77 | 28.6 | |
| CounTX2026.05 | 11.88 | 27.04 | |
| CountTX2026.06 | 11.88 | 27.04 | |
| ZSC (Swin-T)Backbone=Swin-T2026.05 | 12.96 | 26.74 | |
| ZSCBackbone=Swin-T2026.06 | 12.96 | 26.74 | |
| Mean2026.05 | 14.28 | 27.75 | |
| Mean2026.06 | 14.28 | 27.75 | |
| ZSC (ResNet-50)Backbone=ResNet-502026.05 | 14.84 | 31.3 | |
| ZSCBackbone=ResNet-502026.06 | 14.84 | 31.3 | |
| TFOC2026.05 | 16.08 | 31.61 | |
| Qwen3.5-9BModel size=9B2026.05 | 27.4 | 103.5 | |
| Qwen3-VL-8BModel size=8B2026.05 | 31.8 | 142.5 |