Downloads · 30 days
0
Pranilllllll/geonusaf-segformer-b0-legacy-legacy_kfold-fold1
geonusaf-segformer-b0-legacy-legacy_kfold-fold1 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. It is set up for transformers.
Reproduction of the pre-protocol training script (flat AdamW 3e-5, 300 epochs, plain weighted CE, 7 classes including unlabelled). Not a protocol run - do not mix these numbers into the benchmark table without reading…
Downloads · 30 days
0
Access
Public
Updated Sep 11, 2026
Repo size
2.7 GB
Likes
0
Public
Click a slice to open those files.
.pt59.8 MB · 94%
From the Hugging Face model README
Reproduction of the pre-protocol training script (flat AdamW 3e-5, 300 epochs, plain weighted CE, 7 classes including unlabelled). Not a protocol run - do not mix these numbers into the benchmark table without reading the split and metric fields below.
| field | value |
|---|---|
| architecture | SegformerForSemanticSegmentation, nvidia/segformer-b0-finetuned-ade-512-512 |
| split mode | legacy_kfold (3-fold, seed 42) |
| fold | 1 |
| VARIANT LEAKAGE | 375/408 val images (91.9%) have a mirror in train |
| classes | 7 (Residential, Road, River, Forest, UnusedLand, Agricultural, Unlabelled) |
| class weights | legacy_inv_freq, counted over all, cap None |
| loss | weighted CE only (lovasz 0.0, smoothing 0.0) |
| optimizer | AdamW single group, lr 3e-05 flat, wd 0.01 |
| scheduler / clip / EMA | None / None / False |
| epochs / patience / batch | 300 / 25 / 16 |
| val transform | train transform (legacy) |
| best epoch | 188 |
| val mIoU (legacy per-batch, 7-class) | 0.5891 |
| val mIoU (pooled, 7-class) | 0.6300 |
| val mIoU (pooled, 6 real classes) | 0.7193 |
| val OA | 0.8895 |
| val kappa | 0.8328 |
| class | IoU | F1 |
|---|---|---|
| Residential | 0.8470 | 0.9172 |
| Road | 0.4542 | 0.6246 |
| River | 0.5144 | 0.6793 |
| Forest | 0.8994 | 0.9471 |
| UnusedLand | 0.7849 | 0.8795 |
| Agricultural | 0.8161 | 0.8987 |
| Unlabelled | 0.0942 | 0.1723 |