Watermark Remover - Segmentation Model
UNet++ segmentation model for detecting watermarks in images.
Model Details
- Architecture: UNet++ with EfficientNet-B4 encoder
- Framework: PyTorch + segmentation-models-pytorch
- Training: Modal GPU (T4)
- Current checkpoint: Banana/Gemini provider (epoch 48, val_iou=97.44%)
Usage
import torch
import segmentation_models_pytorch as smp
from huggingface_hub import hf_hub_download
# Download weights
weights_path = hf_hub_download(
repo_id="christophernavas/watermark-remover",
filename="segmenter.pth"
)
# Load model
model = smp.UnetPlusPlus(
encoder_name="efficientnet-b4",
encoder_weights=None,
in_channels=3,
classes=1,
activation=None,
)
model.load_state_dict(torch.load(weights_path, map_location="cpu"))
model.eval()
Performance
| Provider | Detection Rate | val_iou |
|---|---|---|
| Banana/Gemini | 95% (20/21) | 97.44% |
License
MIT