
C-FCN PyTorch Implementation
This is a PyTorch implementation of C-FCN, a low power convolutional neural network for cloud segmentation in satellite images, as proposed in "Low-power neural networks for semantic segmentation of satellite images" (Balh et al., 2019).
In addition to implementing the original C-FCN architecture, this implementation includes an option to change the final layer to potentially improve accuracy. The available options for the final layer are:
- Bilinear Upscaling (original)
- 2D Transpose Convolution (scale factor: 4)
- 2x 2D Transpose Convolution (scale factor: 2)
This implementation also includes a skipped connection.
GitHub: https://github.com/hdmquan/C-FCN-PyTorch-implementation