Implementaion of the adaptive-scale U-Net with attention gate (AG), bottleneck attention module (BAM), convolutional block attention module (CBAM), squeeze-and-exitation (SE) and selective-kernel (SK).
AG: https://arxiv.org/abs/1808.08114
BAM: https://arxiv.org/abs/1807.06514
CBAM: https://arxiv.org/abs/1807.06521
SE: https://arxiv.org/abs/1709.01507?spm=a2c41.13233144.0.0
SK: https://arxiv.org/abs/1903.06586
Support multi-head prediction, shortcut connection, dynamic feature selection.
Support cross-entropy loss, strutural similarity (SSIM) loss, Dice loss and IoU loss.
Tested on datasets from PAIP 2019 https://paip2019.grand-challenge.org/Dataset/ and PAIP 2020 https://paip2020.grand-challenge.org/Dataset/.
