Zones segmentation on CaFFe (test)
0.836Overall IoUTyrion-T-GRU
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Tyrion-T-GRUEnsembling=true2025.12 | 0.836 | — | 0.955 | 0.683 | 0.783 | 0.923 | |
| Tyrion-T-LTAEEnsembling=true2025.12 | 0.831 | — | 0.95 | 0.673 | 0.777 | 0.926 | |
| Tyrion-T-GRUEnsembling=false2025.12 | 0.821 | — | 0.951 | 0.653 | 0.765 | 0.916 | |
| Tyrion-T-ConvEnsembling=true2025.12 | 0.821 | — | 0.943 | 0.647 | 0.771 | 0.924 | |
| Tyrion-T-LTAEEnsembling=false2025.12 | 0.818 | — | 0.946 | 0.649 | 0.763 | 0.915 | |
| Tyrion-T-ConvEnsembling=false2025.12 | 0.812 | — | 0.938 | 0.632 | 0.762 | 0.916 | |
| Tyrion-TEnsembling=true2025.12 | 0.798 | — | 0.942 | 0.593 | 0.747 | 0.91 | |
| Tyrion-TEnsembling=false2025.12 | 0.787 | — | 0.936 | 0.584 | 0.737 | 0.889 | |
| TyrionEnsembling=false2025.12 | 0.779 | — | 0.936 | 0.592 | 0.731 | 0.855 | |
| Vincent et al.Ensembling=false2025.12 | 0.561 | — | 0.848 | 0.439 | 0.565 | 0.392 | |
| AMD-HookNetArchitecture Type=CNN, Parameters=14.9M, FLOPs=45.7G, Training Memory=30.2GB, Inference Throughput=182.1 Image/s2025.12 | — | 0.744 | — | — | — | — | |
| AMD-HookNet++Architecture Type=Hybrid, Parameters=48.6M, FLOPs=22.0G, Training Memory=78.3GB, Inference Throughput=164.7 Image/s2025.12 | — | 0.782 | — | — | — | — | |
| BaselineArchitecture Type=CNN, Parameters=19.3M, FLOPs=16.6G, Training Memory=6.9GB, Inference Throughput=577.8 Image/s2025.12 | — | 0.697 | — | — | — | — | |
| HED-U-NetArchitecture Type=CNN, Parameters=8.1M, FLOPs=37.9G, Training Memory=11.5GB, Inference Throughput=195.7 Image/s2025.12 | — | 0.696 | — | — | — | — | |
| HookFormerArchitecture Type=Transformer, Parameters=59.3M, FLOPs=15.4G, Training Memory=70.9GB, Inference Throughput=90.9 Image/s2025.12 | — | 0.755 | — | — | — | — | |
| MISSFormerArchitecture Type=Transformer, Parameters=42.5M, FLOPs=9.9G, Training Memory=12.8GB, Inference Throughput=173.5 Image/s2025.12 | — | 0.644 | — | — | — | — | |
| Swin-UnetArchitecture Type=Transformer, Parameters=41.4M, FLOPs=8.8G, Training Memory=8.3GB, Inference Throughput=215.9 Image/s2025.12 | — | 0.689 | — | — | — | — | |
| Trans-UnetArchitecture Type=Hybrid, Parameters=105.3M, FLOPs=29.3G, Training Memory=10.4GB, Inference Throughput=172.6 Image/s2025.12 | — | 0.66 | — | — | — | — |