[11/2024], pre-trained weights and evalutation code on MUSES are released.
[11/2024], draft paper is available at pdf.
git clone --
cd --
conda create -n anyseg python=3.9
conda activate anyseg
pip install -r requirements.txt
Used Datasets: MUSES / DELIVER
| Method | F | E | L | FE | FL | EL | FEL | Mean | Weights |
|---|---|---|---|---|---|---|---|---|---|
| CMX | 2.52 | 2.35 | 3.01 | 41.15 | 41.25 | 2.56 | 42.27 | 19.30 | - |
| CMNeXt | 3.50 | 2.77 | 2.64 | 6.63 | 10.28 | 3.14 | 46.66 | 10.80 | - |
| MAGIC | 43.22 | 2.68 | 22.95 | 43.51 | 49.05 | 22.98 | 49.02 | 33.34 | - |
| Any2Seg | 44.40 | 3.17 | 22.33 | 44.51 | 49.96 | 22.63 | 50.00 | 33.86 | - |
| Ours | 46.01 | 19.57 | 32.13 | 46.29 | 51.25 | 35.21 | 51.14 | 40.23 | model |
We appreciate the previous open-source works: DELIVER / SegFormer / MUSES