metadata
license: apache-2.0
library_name: pytorch
tags:
- remote-sensing
- disaster-mapping
- oil-spill
- sar
- segmentation
unet.pth — AnyDisasterMapping demo checkpoint (oil_spill)
Task-checkpoint for the AnyDisasterMapping model zoo (unet key, oil-spill SAR segmentation, 3-class).
- Architecture: UNet, 28.96M params
- Training: 3 epochs on tiny synthetic smoke tiles (64x64, 8 samples), CPU smoke run
- Test metrics (synthetic split): OA 5.10% / mIoU 4.18% — demo / pipeline-plumbing quality, NOT production accuracy
- Purpose: end-to-end validation of the weights distribution pipeline. Replace with real trained weights later by uploading a new unet.pth + refreshed SHA256SUMS.
Download
python tools/download_weights.py --list
python tools/download_weights.py --model unet
Checksums in SHA256SUMS (sha256sum format), auto-verified on download.