Downloads · 30 days
0
geonusaf-segformer-b0-loveda-both-v3-slide
Pranilllllll/geonusaf-segformer-b0-loveda-both-v3-slide
geonusaf-segformer-b0-loveda-both-v3-slide is a machine learning model from Pranilllllll. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
LoveDA land-cover segmentation, 7 classes, ignoreindex=255, 512x512 crops @ 0.3 m/px, evaluated at 1024x1024 on the official validation split.
Access
Public
Updated Aug 30, 2026
Repo size
644 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pt74.7 MB · 97%
At a glance
- Access
- Public
- Created
- Aug 30, 2026
- Updated
- Aug 30, 2026
- SHA
- 2f4610f8
Datasets
Model card
From the Hugging Face model README
GeoNUSAF segformer-b0 -- LoveDA benchmark (both)
LoveDA land-cover segmentation, 7 classes, ignore_index=255, 512x512 crops @ 0.3 m/px, evaluated at 1024x1024 on the official validation split.
| backbone | nvidia/segformer-b0-finetuned-ade-512-512 (ADE20K init only) |
| split | official LoveDA Train -> Val |
| budget | 20000 steps, batch 16, single seed 42 |
| best step | 9000 |
| detail path | False |
| CSA reweighting | False (tau=[0.6, 0.6, 0.35, 0.35, 0.6, 0.6, 0.6], w_min=0.5) |
| soft-clDice | False (mu=0.3, classes Road/Water) |
| balanced sampler | False |
Not a leaderboard entry. This is a fixed-budget controlled comparison against the SegFormer-B0 baseline trained identically, not the 80k-iteration schedule used by published LoveDA results.
Validation (official Val split)
mIoU 0.4730 | mF1 0.6369 | OA 0.6502 | kappa 0.5583
| class | IoU | UA (prec) | PA (rec) |
|---|---|---|---|
| Background | 0.4436 | 0.6370 | 0.5937 |
| Building | 0.5051 | 0.5462 | 0.8704 |
| Road | 0.5310 | 0.6696 | 0.7196 |
| Water | 0.6038 | 0.6526 | 0.8897 |
| Barren | 0.3149 | 0.5689 | 0.4136 |
| Forest | 0.3870 | 0.4522 | 0.7285 |
| Agriculture | 0.5255 | 0.8743 | 0.5685 |
- Created
- Aug 30, 2026
- Updated
- Aug 30, 2026