u-rwkv-busi / README.md
hbyecoding's picture
Initial commit
3cb213d
metadata
language: en
tags:
  - medical-image-segmentation
  - pytorch
  - u-rwkv
datasets:
  - rwkv
metrics:
  - dice
model-index:
  - name: u-rwkv-polyp
    results:
      - task:
          type: image-segmentation
          name: Medical Image Segmentation
        dataset:
          type: rwkv
          name: RWKV
        metrics:
          - type: dice
            value: 0.7887

u-rwkv-polyp

This model is part of the U-RWKV family of medical image segmentation models. It combines the power of RWKV (Receptance Weighted Key Value) attention mechanism with U-Net architecture for efficient and accurate medical image segmentation.

Model description

U-RWKV model trained on Polyp dataset for polyp segmentation

Architecture

  • Base architecture: U-Net with RWKV attention
  • Input channels: 3
  • Output channels: 1
  • Base channels: [16, 32, 128, 160, 256]
  • Attention mechanism: RWKV (Receptance Weighted Key Value)
  • Feature fusion: SE (Squeeze-and-Excitation)

Performance

  • Dice score: 0.7887

Usage

import torch
from models.model import U_RWKV

# Load model
model = U_RWKV()
checkpoint = torch.load('model.pth')
model.load_state_dict(checkpoint['model_state_dict'])
model.eval()

# Inference
with torch.no_grad():
    output = model(input_image)

Training

The model was trained using:

  • Loss functions: Dice Loss + BCE Loss
  • Optimizer: AdamW
  • Learning rate: 1e-4
  • Batch size: 16
  • Data augmentation: Random flip, rotation, scaling

License

This model is released under the MIT License.