Visual Grounding on RefExp
97AccuracyPLC
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| PLCInput type=Pipelined2022.10 | 97 | — | |
| VUTInput type=Pipelined, Android view hierarchy=true2022.10 | 94.8 | — | |
| Pix2Struct-LargeInput type=Pixel-only, Scale=Large2022.10 | 94.2 | — | |
| Pix2Struct-BaseInput type=Pixel-only, Scale=Base2022.10 | 92.2 | — | |
| UI BertInput type=Pipelined, Android view hierarchy=true2022.10 | 90.8 | — | |
| AguvisSize=8B, TTFT ↓ (ms)=119.7, TPOT ↓ (ms/token)=21.22026.04 | 80.9 | — | |
| GoClick-LSize=0.8B, TTFT ↓ (ms)=91.1, TPOT ↓ (ms/token)=8.32026.04 | 78.2 | — | |
| OS-ATLAS-4BSize=4B, TTFT ↓ (ms)=137.3, TPOT ↓ (ms/token)=31.42026.04 | 77.1 | — | |
| UGroundSize=7B, TTFT ↓ (ms)=1034.6, TPOT ↓ (ms/token)=27.92026.04 | 73.6 | — | |
| GoClick-BSize=0.2B, TTFT ↓ (ms)=37.7, TPOT ↓ (ms/token)=4.12026.04 | 71.9 | — | |
| OS-ATLAS-8BSize=8B, TTFT ↓ (ms)=137.5, TPOT ↓ (ms/token)=19.92026.04 | 66.5 | — | |
| Qwen2VL-7BSize=8B, TTFT ↓ (ms)=118.9, TPOT ↓ (ms/token)=21.22026.04 | 64.8 | — | |
| ShowUISize=2B, TTFT ↓ (ms)=79.7, TPOT ↓ (ms/token)=14.72026.04 | 58.4 | — | |
| SeeClickSize=10B, TTFT ↓ (ms)=160.4, TPOT ↓ (ms/token)=184.42026.04 | 58.1 | — | |
| CogAgentSize=18B, TTFT ↓ (ms)=1253.2, TPOT ↓ (ms/token)=208.82026.04 | 35 | — | |
| Qwen2-VLSize=2B, TTFT ↓ (ms)=58.8, TPOT ↓ (ms/token)=16.42026.04 | 29.2 | — | |
| GPT-4o2026.04 | 21.8 | — | |
| Ferret-UISize=8B, TTFT ↓ (ms)=152.5, TPOT ↓ (ms/token)=22.92026.04 | 5.5 | — | |
| Ferret-UISize=3B, TTFT ↓ (ms)=69.5, TPOT ↓ (ms/token)=9.82026.04 | 1.1 | — | |
| CLIP2025.12 | — | 13.15 | |
| GeoRSCLIP2025.12 | — | 8.25 | |
| RemoteCLIP2025.12 | — | 4.35 | |
| SkyCLIP2025.12 | — | 11.85 | |
| VLM2GeoVecEvaluation Protocol=In-distribution2025.12 | — | 32.5 | |
| VLM2Vec2025.12 | — | 6.65 |