Semantic Segmentation on ADE20K 100-10 overlapped (val)
26.1mIoU (101-110)Joint
Evaluation results
| Method | Links | |||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| JointContinual Learning Scenario=100-10, Steps=6 steps2022.03 | 26.1 | 44.3 | 42.8 | 26.7 | 28.1 | 17.3 | 38.9 | — | — | — | — | — | — | — | — | — | — | |
| PLOPContinual Learning Scenario=100-10, Steps=6 steps2022.03 | 15.2 | 40.6 | 16.9 | 18.7 | 11.9 | 7.9 | 31.6 | — | — | — | — | — | — | — | — | — | — | |
| RCIL (Ours)Continual Learning Scenario=100-10, Steps=6 steps2022.03 | 14.6 | 39.3 | 26.3 | 23.2 | 12.1 | 11.8 | 32.1 | — | — | — | — | — | — | — | — | — | — | |
| MiBContinual Learning Scenario=100-10, Steps=6 steps2022.03 | 12.6 | 38.3 | 10.6 | 8.7 | 9.5 | 15.1 | 29.2 | — | — | — | — | — | — | — | — | — | — | |
| FBLFramework=Federated Learning (FL), Backbone=Deeplab-v3 with ResNet-101, Pre-training=ImageNet, Setting=100-10 with overlapped foregrounds2023.04 | 4.3 | — | 3.1 | 3.3 | 2.7 | 0.2 | 2,790 | 6,750 | 4,790 | 4,890 | 3,880 | 3,320 | 4,260 | 3,540 | 3,520 | 3,280 | 1,790 | |
| MiB + FLFramework=Federated Learning (FL), Backbone=Deeplab-v3 with ResNet-101, Pre-training=ImageNet, Setting=100-10 with overlapped foregrounds2023.04 | 2.8 | — | 5.2 | 5.8 | 6 | 15.4 | 2,540 | 5,970 | 4,130 | 4,240 | 3,240 | 2,790 | 3,640 | 2,870 | 2,890 | 3,010 | 1,470 | |
| PLOP + FLFramework=Federated Learning (FL), Backbone=Deeplab-v3 with ResNet-101, Pre-training=ImageNet, Setting=100-10 with overlapped foregrounds2023.04 | 2.4 | — | 5.2 | 6.8 | 2.5 | 11.5 | 2,610 | 6,070 | 4,340 | 4,380 | 3,370 | 2,880 | 3,700 | 3,110 | 3,030 | 3,210 | 1,590 | |
| RCIL + FLFramework=Federated Learning (FL), Backbone=Deeplab-v3 with ResNet-101, Pre-training=ImageNet, Setting=100-10 with overlapped foregrounds2023.04 | 0.3 | — | 12.3 | 6.3 | 2.6 | 3.5 | 2,640 | 6,300 | 4,660 | 4,720 | 3,510 | 3,120 | 3,600 | 3,080 | 3,270 | 2,810 | 1,620 | |
| ILTContinual Learning Scenario=100-10, Steps=6 steps2022.03 | 0 | 0.1 | 0.1 | 0.9 | 4.1 | 9.3 | 1.1 | — | — | — | — | — | — | — | — | — | — | |
| Finetuning + FLFramework=Federated Learning (FL), Backbone=Deeplab-v3 with ResNet-101, Pre-training=ImageNet, Setting=100-10 with overlapped foregrounds2023.04 | 0 | — | 0 | 0 | 0 | 10.3 | 70 | 80 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | |
| LWF + FLFramework=Federated Learning (FL), Backbone=Deeplab-v3 with ResNet-101, Pre-training=ImageNet, Setting=100-10 with overlapped foregrounds2023.04 | 0 | — | 0.1 | 0 | 1.6 | 9.4 | 90 | 80 | 0 | 0 | 0 | 0 | 50 | 130 | 10 | 0 | 10 | |
| ILT + FLFramework=Federated Learning (FL), Backbone=Deeplab-v3 with ResNet-101, Pre-training=ImageNet, Setting=100-10 with overlapped foregrounds2023.04 | 0 | — | 0.1 | 0 | 0.7 | 4.3 | 60 | 80 | 0 | 0 | 30 | 0 | 50 | 190 | 20 | 10 | 30 |