... and I stopped tracking scores here at 0.33915 :( but best public LB score in the repo is 0.270
Model (.scripts/ folder) |
Image Size | Epochs | Bag | TTA | Fold | Val | LB | Comment |
|---|---|---|---|---|---|---|---|---|
| 2XSPPSeresnext50; 2XTSM Resnet50 | 224 | 16,17; 4,5 | 1 | Hflip | 0 | 0.168, 0.163; 0.158, 0.196 | 0.33915 | Cosine restart at 10; ratio clip; BSize 4; spp15 single lstm 256 HU tsm01 TSM with avg consesnus |
| 2XSPPSeresnext50; 2XTSM Resnet50 | 224 | 16,17; 4,5 | 1 | None | 0 | 0.168, 0.163; 0.158, 0.196 | 0.34033 | Cosine restart at 10; ratio clip; BSize 4; spp15 single lstm 256 HU tsm01 TSM with avg consesnus |
| 2XSPPResnet50; 2XTSM Resnet50 | 224 | 6,8; 4,5 | 1 | Hflip | 0 | 0.183, 0.174; 0.158, 0.196 | 0.34486 | Change clip to ratio clip; spp14 single lstm 256 HU tsm01 TSM with avg consesnus |
| 2XSPPResnet50; 2XTSM Resnet50 | 224 | 6,8; 4,5 | 1 | Hflip | 0 | 0.183, 0.174; 0.158, 0.196 | 0.34683 | spp14 single lstm 256 HU tsm01 TSM with avg consesnus |
| 2XSPPResnet50; 2XTSM Resnet50 | 224 | 6,8; 4,5 | 1 | NA | 0 | 0.183, 0.174; 0.158, 0.196 | 0.35405 | spp14 single lstm 256 HU tsm01 TSM with avg consesnus |
| Resnet50 and one 34; SPPNet | 224 | 7,8,9 | 3 | NA | 0 | ~0.174 (0.183, 0.182, 0.174) | 0.362 | spp14 single lstm 256 HU |
| Resnet34; SPPNet | 224 | 7,8,9 | 3 | NA | 0 | 0.1786 (0.179, 0.219, 0.210) | 0.386 | spp13 single lstm 256 HU |