Egocentric Spatial Reasoning on 3DSRBench Egocentric (test)
57.43Orientation Accuracy (Cam.V)Alloceiver + Qwen2.5VL-32B
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Alloceiver + Qwen2.5VL-32BBackbone VLM=Qwen2.5VL-32B, Enhancement (Alloceiver)=true2026.02 | 57.43 | 54.03 | 44.24 | 69.29 | 79.87 | 79.35 | 61.72 | |
| Alloceiver + InternVL2.5-38BBackbone VLM=InternVL2.5-38B, Enhancement (Alloceiver)=true2026.02 | 54.52 | 55.89 | 50.29 | 60.57 | 84.81 | 80.53 | 63.18 | |
| Alloceiver + Qwen2.5VL-7BBackbone VLM=Qwen2.5VL-7B, Enhancement (Alloceiver)=true2026.02 | 35.14 | 42.92 | 46.5 | 62.14 | 82.3 | 65.63 | 54.47 | |
| InternVL2.5-38BBackbone VLM=InternVL2.5-38B, Enhancement (Alloceiver)=false2026.02 | 34.94 | 53.98 | 47.08 | 65 | 83.04 | 76.4 | 58.91 | |
| Qwen2.5VL-32BBackbone VLM=Qwen2.5VL-32B, Enhancement (Alloceiver)=false2026.02 | 29.92 | 46.02 | 43.44 | 66.71 | 79.87 | 72.42 | 54.56 | |
| Qwen2.5VL-7BBackbone VLM=Qwen2.5VL-7B, Enhancement (Alloceiver)=false2026.02 | 27.81 | 45.4 | 42.86 | 58.71 | 81.49 | 60.03 | 52.26 | |
| Alloceiver + GPT-4oBackbone VLM=GPT-4o, Enhancement (Alloceiver)=true2026.02 | 26.41 | 58.88 | 56.71 | 71.29 | 82.45 | 75.37 | 60.55 | |
| Random2026.02 | 25 | 25 | 25 | 50 | 50 | 50 | 34.94 | |
| GPT-4oBackbone VLM=GPT-4o, Enhancement (Alloceiver)=false2026.02 | 19.78 | 50.48 | 47.16 | 67.86 | 79.72 | 74.48 | 52.27 | |
| Alloceiver + Qwen2.5VL-32Btype=Training-free, backbone=Qwen2.5VL-32B2026.02 | 0.5743 | 0.5403 | 0.4424 | 0.6929 | 0.7987 | 0.7935 | 0.6172 | |
| Alloceiver + InternVL2.5-38Btype=Training-free, backbone=InternVL2.5-38B2026.02 | 0.5452 | 0.5589 | 0.5029 | 0.6057 | 0.8481 | 0.8053 | 0.6318 | |
| RoboBrain2.0-32Btype=Trained Spatial VLM2026.02 | 0.3665 | 0.489 | 0.4512 | 0.7429 | 0.8577 | 0.8068 | 0.5936 | |
| Alloceiver + Qwen2.5VL-7Btype=Training-free, backbone=Qwen2.5VL-7B2026.02 | 0.3514 | 0.4292 | 0.465 | 0.6214 | 0.823 | 0.6563 | 0.5447 | |
| InternVL2.5-38Btype=Vanilla VLM2026.02 | 0.3494 | 0.5398 | 0.4708 | 0.65 | 0.8304 | 0.764 | 0.5891 | |
| Gemini-2.5-protype=Vanilla VLM2026.02 | 0.3333 | 0.5764 | 0.5175 | 0.7514 | 0.8282 | 0.8392 | 0.6228 | |
| Spatial-SSRL-7Btype=Trained Spatial VLM2026.02 | 0.3233 | 0.454 | 0.4118 | 0.65 | 0.84 | 0.6504 | 0.542 | |
| SpatialThinker-7Btype=Trained Spatial VLM2026.02 | 0.3052 | 0.4456 | 0.4271 | 0.59 | 0.8193 | 0.5678 | 0.522 | |
| REVPT-7Btype=Trained Spatial VLM2026.02 | 0.2962 | 0.4365 | 0.4031 | 0.6357 | 0.8208 | 0.615 | 0.4969 | |
| GPT-5-minitype=Vanilla VLM2026.02 | 0.2701 | 0.5753 | 0.5394 | 0.7543 | 0.8525 | 0.7227 | 0.6113 | |
| Alloceiver + GPT-4otype=Training-free, backbone=GPT-4o2026.02 | 0.2641 | 0.5888 | 0.5671 | 0.7129 | 0.8245 | 0.7537 | 0.6055 | |
| Randomtype=Baseline2026.02 | 0.25 | 0.25 | 0.25 | 0.5 | 0.5 | 0.5 | 0.3494 | |
| Cosmos-Reason1-7Btype=Trained Spatial VLM2026.02 | 0.2359 | 0.4264 | 0.406 | 0.5671 | 0.7788 | 0.5944 | 0.4951 | |
| SenseNova-SI-1.1-7Btype=Trained Spatial VLM2026.02 | 0.2038 | 0.4569 | 0.3025 | 0.6586 | 0.8164 | 0.5369 | 0.4887 | |
| GPT-4otype=Vanilla VLM2026.02 | 0.1978 | 0.5048 | 0.4716 | 0.6786 | 0.7972 | 0.7448 | 0.5227 |