4D Panoptic Segmentation on SemanticKITTI (test)
68.4LSTQMask4Former
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Mask4Former2023.09 | 68.4 | 67.3 | 69.6 | 72.7 | 65.3 | |
| Eq-4D-StOPc (width)=128, n (anchor sizes)=42023.03 | 67.8 | 72.3 | 63.5 | 70.4 | 61.9 | |
| Eq-4D-StOP*Concurrent Work=true2023.09 | 67.8 | 72.3 | 63.5 | 70.4 | 61.9 | |
| Mask4D*Concurrent Work=true2023.09 | 64.3 | 66.4 | 62.2 | 69.9 | 52.2 | |
| 4D-StOPTraining=End-to-End, Number of scans=42022.09 | 63.9 | 69.5 | 58.8 | 67.7 | 53.8 | |
| 4D-StOP2023.03 | 63.9 | 69.5 | 58.8 | 67.7 | 53.8 | |
| 4D-StOP2023.09 | 63.9 | 69.5 | 58.8 | 67.7 | 53.8 | |
| CIATraining=Non-End-to-End2022.09 | 63.1 | 65.7 | 60.6 | 66.9 | 52 | |
| CA-Net2023.03 | 63.1 | 65.7 | 60.6 | 66.9 | 52 | |
| CIA2023.09 | 63.1 | 65.7 | 60.6 | 66.9 | 52 | |
| 4D-StOPTraining=End-to-End, Number of scans=22022.09 | 62.9 | 67.3 | 58.8 | 68.3 | 53.3 | |
| 4D-DS-Net2023.03 | 62.3 | 65.8 | 58.9 | 65.6 | 49.8 | |
| 4D-DS-Net2023.09 | 62.3 | 65.8 | 58.9 | 65.6 | 49.8 | |
| 4D-PLSTraining=End-to-End, Number of scans=42022.09 | 56.9 | 56.4 | 57.4 | 66.9 | 51.6 | |
| 4D-PLS2023.03 | 56.9 | 56.4 | 57.4 | 66.9 | 51.6 | |
| 4D-PLS2023.09 | 56.9 | 56.4 | 57.4 | 66.9 | 51.6 | |
| KPConv + PP + SFPTraining=Non-End-to-End, Components=PointPillars, scene flow based propagation2022.09 | 38.5 | 26.6 | 55.9 | 66.9 | 47.7 | |
| KPConvPost-processing=PP+SFP2023.03 | 38.5 | 26.6 | 55.9 | 66.9 | 47.7 | |
| KPConv + PP + SFPDetection/Segmentation=PointPillars (PP), Propagation=Scene flow based propagation (SFP)2023.09 | 38.5 | 26.6 | 55.9 | 66.9 | 47.7 | |
| KPConv + PP + MOTTraining=Non-End-to-End, Components=PointPillars, tracking-by-detection2022.09 | 38 | 25.9 | 55.9 | 66.9 | 47.7 | |
| KPConvPost-processing=PP+MOT2023.03 | 38 | 25.9 | 55.9 | 66.9 | 47.7 | |
| KPConv + PP + MOTDetection/Segmentation=PointPillars (PP), Tracking=Multi-Object Tracking (MOT)2023.09 | 38 | 25.9 | 55.9 | 66.9 | 47.7 | |
| RangeNet++ + PP + MOTTraining=Non-End-to-End, Components=PointPillars, tracking-by-detection2022.09 | 35.5 | 24.1 | 52.4 | 64.5 | 35.8 | |
| RangeNet++Post-processing=PP+MOT2023.03 | 35.5 | 24.1 | 52.4 | 64.5 | 35.8 | |
| RangeNet++ + PP + SFPTraining=Non-End-to-End, Components=PointPillars, scene flow based propagation2022.09 | 34.9 | 23.3 | 52.4 | 64.5 | 35.8 | |
| RangeNet++Post-processing=PP+SFP2023.03 | 34.9 | 23.3 | 52.4 | 64.5 | 35.8 | |
| RangeNet++ + PP + SFPDetection/Segmentation=PointPillars (PP), Propagation=Scene flow based propagation (SFP)2023.09 | 34.9 | 23.3 | 52.4 | 64.5 | 35.8 |