├── Active-Contour-Loss.py
├── README.md
└── notebooks
    └── demo.ipynb


/Active-Contour-Loss.py:
--------------------------------------------------------------------------------
 1 | 
 2 | import torch
 3 | 
 4 | def active_contour_loss(y_true, y_pred, weight=10):
 5 |   '''
 6 |   y_true, y_pred: tensor of shape (B, C, H, W), where y_true[:,:,region_in_contour] == 1, y_true[:,:,region_out_contour] == 0.
 7 |   weight: scalar, length term weight.
 8 |   '''
 9 |   # length term
10 |   delta_r = y_pred[:,:,1:,:] - y_pred[:,:,:-1,:] # horizontal gradient (B, C, H-1, W) 
11 |   delta_c = y_pred[:,:,:,1:] - y_pred[:,:,:,:-1] # vertical gradient   (B, C, H,   W-1)
12 |   
13 |   delta_r    = delta_r[:,:,1:,:-2]**2  # (B, C, H-2, W-2)
14 |   delta_c    = delta_c[:,:,:-2,1:]**2  # (B, C, H-2, W-2)
15 |   delta_pred = torch.abs(delta_r + delta_c) 
16 | 
17 |   epsilon = 1e-8 # where is a parameter to avoid square root is zero in practice.
18 |   lenth = torch.mean(torch.sqrt(delta_pred + epsilon)) # eq.(11) in the paper, mean is used instead of sum.
19 |   
20 |   # region term
21 |   c_in  = torch.ones_like(y_pred)
22 |   c_out = torch.zeros_like(y_pred)
23 | 
24 |   region_in  = torch.mean( y_pred     * (y_true - C_in )**2 ) # equ.(12) in the paper, mean is used instead of sum.
25 |   region_out = torch.mean( (1-y_pred) * (y_true - C_out)**2 ) 
26 |   region = region_in + region_out
27 |   
28 |   loss =  weight*lenth + region
29 | 
30 |   return loss
31 | 


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
1 | # Active-Contour-Loss-pytorch
2 | An unofficial pytorch implementation for cvpr2019 paper "Learning Active Contour Models for Medical Image Segmentation" by Chen, Xu, et al.
3 | 
4 | The offcial tensorflow implementation can be found at https://github.com/xuuuuuuchen/Active-Contour-Loss
5 | 


--------------------------------------------------------------------------------