3D Sorting on 3D Sorting
92.4mIoUQwen3-8B (Ours)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen3-8B (Ours)Model Category=Reasoning Models, Protocol=Ours2026.03 | 92.4 | 0.145 | 100 | 0.51 | |
| Qwen2.5-7B (Ours)Model Category=Instruct Models, Protocol=Ours2026.03 | 87.9 | 0.179 | 100 | 0.627 | |
| Cosmos-7B (Ours)Model Category=Reasoning Models, Protocol=Ours2026.03 | 87.5 | 0.234 | 100 | 0.592 | |
| Qwen3 (Vanilla GRPO)Model Category=Reasoning Models, Protocol=Vanilla GRPO2026.03 | 85 | 0.274 | 95.4 | 0.684 | |
| Qwen2.5-VL-7B (Ours)Model Category=Instruct Models, Protocol=Ours2026.03 | 79.8 | 0.228 | 96.1 | 0.613 | |
| LayoutGPT (Gemini-2.5)Model Category=Large Language Models, Protocol=Zero-shot, Backbone=Gemini-2.52026.03 | 78.1 | 0.471 | 98.2 | 0.709 | |
| Deepseek-R1Model Category=Large Language Models, Protocol=Zero-shot2026.03 | 77.6 | 0.566 | 90.3 | 0.783 | |
| LayoutGPT (Qwen3-235B)Model Category=Large Language Models, Protocol=Zero-shot, Backbone=Qwen3-235B2026.03 | 76.3 | 0.518 | 97.1 | 0.91 | |
| Gemini 2.5 FlashModel Category=Large Language Models, Protocol=Zero-shot2026.03 | 74.6 | 0.458 | 96.8 | 1.84 | |
| Cosmos-7B (Vanilla GRPO)Model Category=Reasoning Models, Protocol=Vanilla GRPO2026.03 | 72.9 | 0.297 | 93.7 | 0.811 | |
| Gemini 2.5 ProModel Category=Large Language Models, Protocol=Zero-shot2026.03 | 72.4 | 0.593 | 97.6 | 2.18 | |
| Qwen2.5-VL-7B (CoT-SFT)Model Category=Instruct Models, Protocol=CoT-SFT2026.03 | 71.3 | 0.33 | 84.7 | 0.853 | |
| Qwen2.5-7B (CoT-SFT)Model Category=Instruct Models, Protocol=CoT-SFT2026.03 | 71.2 | 0.345 | 89.8 | 0.992 | |
| Qwen3-235B-ThinkingModel Category=Large Language Models, Protocol=Zero-shot2026.03 | 70.8 | 0.856 | 95.7 | 2.208 | |
| Qwen3-VL-235B-ThinkingModel Category=Large Language Models, Protocol=Zero-shot2026.03 | 70.6 | 0.568 | 91.8 | 1.9 | |
| Cosmos-7B (CoT-SFT)Model Category=Reasoning Models, Protocol=CoT-SFT2026.03 | 64.1 | 0.41 | 75.7 | 1.383 | |
| Qwen3-8B (CoT-SFT)Model Category=Reasoning Models, Protocol=CoT-SFT2026.03 | 63.5 | 0.513 | 85.4 | 1.61 | |
| Qwen2.5-7B (Ans-SFT)Model Category=Instruct Models, Protocol=Ans-SFT2026.03 | 62.2 | 0.348 | 74.5 | 1.175 | |
| Deepseek-R1-0528Model Category=Large Language Models, Protocol=Zero-shot2026.03 | 60.1 | 1.163 | 79.9 | 2.021 | |
| Qwen2.5-VL-7B (Ans-SFT)Model Category=Instruct Models, Protocol=Ans-SFT2026.03 | 57 | 0.423 | 69.8 | 0.9 | |
| Qwen3-8B (Ans-SFT)Model Category=Reasoning Models, Protocol=Ans-SFT2026.03 | 55.2 | 0.493 | 70 | 1.13 | |
| Cosmos-7B (Ans-SFT)Model Category=Reasoning Models, Protocol=Ans-SFT2026.03 | 50.1 | 0.577 | 68.3 | 1.794 | |
| Deepseek-V3Model Category=Large Language Models, Protocol=Zero-shot2026.03 | 23.6 | 1.242 | 62 | 2.01 | |
| Qwen3-VL-235B-InstructModel Category=Large Language Models, Protocol=Zero-shot2026.03 | 14.5 | 2.323 | 42.1 | 8.12 | |
| Qwen3-235B-InstructModel Category=Large Language Models, Protocol=Zero-shot2026.03 | 12.2 | 2.702 | 54.8 | 6.285 |