3D Situated Question Answering on SQA3D
47.7What Accuracy3D-LLM
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| 3D-LLMModality=V, P, Trainable Params.=216M, GFLOPS=1.67K2024.02 | 47.7 | 61 | 49 | 63.6 | 49 | 49.6 | 52.3 | |
| CREMAModality=V, P, D, N, Trainable Params.=38M, GFLOPS=1.83K2024.02 | 47.6 | 66 | 51 | 65.1 | 48.4 | 56.4 | 54.6 | |
| CREMAModality=V, P, Trainable Params.=8M, GFLOPS=1.69K2024.02 | 46.2 | 63.6 | 46.4 | 63 | 48.7 | 50.1 | 52.1 | |
| 3D-LLMModality=V, P, D, N, Trainable Params.=434M, GFLOPS=5.47K2024.02 | 45.2 | 62.4 | 44.8 | 64.2 | 45.8 | 49.4 | 52 | |
| 3D-LLMModality=V, Trainable Params.=108M, GFLOPS=1.30K2024.02 | 45.1 | 62.7 | 48.6 | 63.3 | 45.8 | 49.8 | 51.4 | |
| CREMAModality=V, Trainable Params.=4M, GFLOPS=1.30K2024.02 | 44.9 | 62.1 | 48.1 | 67.7 | 48.4 | 49.8 | 51.8 | |
| ScanQAModality=P (3D Point cloud), Trainable Params.=38M2024.02 | 31.6 | 63.8 | 46.2 | 69.5 | 43.8 | 45.3 | 46.5 | |
| ClipBERTModality=V (Video RGB frames), Trainable Params.=135M2024.02 | 30.2 | 60.1 | 38.7 | 63.3 | 42.4 | 42.7 | 43.3 | |
| MCANModality=V* (Bird-Eye View image), Trainable Params.=56M2024.02 | 28.8 | 59.6 | 44 | 68.3 | 40.7 | 40.4 | 43.4 |