Semantic Segmentation on ScanNet scenes (val)
53.8mIoU (Step 1)Ours (novel viewpoints)
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Ours (novel viewpoints)adaptation_stage=Adapt, replay_strategy=novel viewpoints replay2022.11 | 53.8 | 46.7 | 23.2 | 49 | 42.9 | 40.1 | 58 | 23.2 | 66.7 | 71.5 | 47.5 | |
| Oursadaptation_stage=Adapt, replay_strategy=joint training2022.11 | 53.7 | 46.3 | 24.3 | 49.1 | 43.7 | 40.4 | 55.8 | 26.2 | 68.9 | 72.5 | 48.1 | |
| Ours (Ipre replay only)adaptation_stage=Adapt, replay_strategy=pre-training data replay only2022.11 | 53.3 | 48 | 20.5 | 49 | 43.4 | 39 | 62.1 | 26.7 | 65.7 | 73 | 48.1 | |
| Mappingadaptation_stage=Adapt2022.11 | 46.8 | 42.1 | 23.6 | 50.6 | 44 | 35.8 | 56.7 | 26.5 | 68.3 | 72.7 | 46.7 | |
| CoTTAadaptation_stage=Adapt2022.11 | 44 | 40.9 | 22.7 | 50.2 | 40 | 37.5 | 56 | 26.9 | 54.5 | 73.8 | 44.7 | |
| Pre-trainadaptation=none2022.11 | 43.9 | 41.3 | 23 | 50.2 | 40.1 | 37.6 | 55.8 | 27.9 | 54.9 | 73.5 | 44.8 |