3D Object Classification on Objaverse
72.5Accuracy (Instruction-typed)PointAlign
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| PointAlignLLM Size=2.7B, 3D Data Size=730K, Input=Point Cloud2026.02 | 72.5 | 69.5 | 71 | |
| MiniGPT-3DLLM Size=2.7B, 3D Data Size=730K, Input=Point Cloud2026.02 | 65 | 68.5 | 66.75 | |
| PointLLM-7BLLM Size=7B, 3D Data Size=730K, Input=Point Cloud2026.02 | 62 | 63 | 62.5 | |
| PointLLM-13BLLM Size=13B, 3D Data Size=730K, Input=Point Cloud2026.02 | 61.5 | 63 | 62.25 | |
| MiniGPT-3DLLM Size=2.7B, Trainable Params=0.05B (47.8M), Input=3D Point Cloud2024.05 | 60 | 60.5 | 60.25 | |
| PointLLM-R-7Bzero-shot=true, parameters=7B, protocol=PointLLM2026.05 | 59.17 | 59.27 | — | |
| PointLLM-13BLLM Size=13B, Trainable Params=13.01B, Input=3D Point Cloud2024.05 | 56.5 | 51.5 | 54 | |
| PointLLM-13BInput=3D Point Cloud, Zero-shot=true2023.08 | 56.5 | 51.5 | 53.39 | |
| PointLLM-7BLLM Size=7B, Trainable Params=7.01B, Input=3D Point Cloud2024.05 | 55 | 51 | 53 | |
| PointLLM-7BInput=3D Point Cloud, Zero-shot=true2023.08 | 55 | 51 | 52.82 | |
| LLAVA-13BLLM Size=13B, Trainable Params=13.03B, Input=Single-V. Img.2024.05 | 53 | 50.5 | 51.75 | |
| LLaVA-13BInput=Single-V. Img., Zero-shot=true2023.08 | 53 | 50.5 | 44.17 | |
| MiniGPT-3Dzero-shot=true, protocol=PointLLM2026.05 | 52.9 | 52.63 | — | |
| PointLLM-13Bzero-shot=true, parameters=13B, protocol=PointLLM2026.05 | 49.83 | 49.78 | — | |
| LLAVA-7BLLM Size=7B, Trainable Params=7.03B, Input=Single-V. Img.2024.05 | 49.5 | 50.5 | 50 | |
| LLaVA-7BInput=Single-V. Img., Zero-shot=true2023.08 | 49.5 | 50.5 | 44.86 | |
| 3D-LLMLLM Size=13B, Input=3D Obj. + Mul.-V. Img.2024.05 | 49 | 41.5 | 45.25 | |
| 3D-LLMInput=3D Obj. + Mul.-V. Img., Zero-shot=true2023.08 | 49 | 41.5 | 45.25 | |
| GPT4PointLLM Size=2.7B, 3D Data Size=660K, Input=Point Cloud2026.02 | 49 | 46.5 | 47.75 | |
| PointLLM-7Bzero-shot=true, parameters=7B, protocol=PointLLM2026.05 | 48.85 | 48.4 | — | |
| InstructBLIP-7BLLM Size=7B, Trainable Params=0.20B, Input=Single-V. Img.2024.05 | 45 | 42 | 43.5 | |
| InstructBLIP-7BInput=Single-V. Img., Zero-shot=true2023.08 | 45 | 42 | 34.5 | |
| LLaVA-13BLLM Size=13B, 3D Data Size=0K, Input=Single-V. Img.2026.02 | 39.5 | 35.5 | 37.5 | |
| GPT-4o miniLLM Size=-, 3D Data Size=0K, Input=Single-V. Img.2026.02 | 39 | 35 | 37 | |
| LLaVA-7BLLM Size=7B, 3D Data Size=0K, Input=Single-V. Img.2026.02 | 37.5 | 30 | 33.75 | |
| InstructBLIP-13BLLM Size=13B, Trainable Params=0.20B, Input=Single-V. Img.2024.05 | 37 | 31.5 | 34.25 | |
| InstructBLIP-13BInput=Single-V. Img., Zero-shot=true2023.08 | 37 | 31.5 | 31.47 | |
| ShapeLLM-13Bzero-shot=true, parameters=13B, protocol=PointLLM2026.05 | 32.28 | 33.47 | — | |
| ShapeLLM-7Bzero-shot=true, parameters=7B, protocol=PointLLM2026.05 | 23.44 | 20.56 | — | |
| InstructBLIP-7BLLM Size=7B, 3D Data Size=0K, Input=Single-V. Img.2026.02 | 21.5 | 26 | 23.75 | |
| InstructBLIP-13BLLM Size=13B, 3D Data Size=0K, Input=Single-V. Img.2026.02 | 21.5 | 21.5 | 21.5 | |
| Point-Bind LLMLLM Size=7B, 3D Data Size=-, Input=Point Cloud2026.02 | 7.5 | 7.58 | 7.54 | |
| Point-Bind LLMLLM Size=7B, Input=3D Point Cloud2024.05 | 6 | 4.5 | 5.25 | |
| Point-Bind LLMInput=3D Point Cloud, Zero-shot=true2023.08 | 6 | 4.5 | 25.53 | |
| ShapeLLM-13BLLM Size=13B, Trainable Params=13.04B, Input=3D Point Cloud2024.05 | — | — | 54 | |
| ShapeLLM-7BLLM Size=7B, Trainable Params=7.04B, Input=3D Point Cloud2024.05 | — | — | 54.5 |