This repo was constructed for cell instance segmentation.
- run horizontalVerticalMap.py to save the horizontal and vertical maps.
- run train.py to train the model
- run inference.py to inference
There are 2 branches consisting of horizontalVertical(MSE loss) and segmentation(Dice + CE)
- MSE loss: (l1)the loss of the horizontal and vertical map + (l2)the loss of the gradient of the horizontal and vertical map
- Dice loss: sementic segmentation output(l3)
- CE loss: sementic segmentation output(l4)
Loss = l1 + 2 * l2 + l3 + l4
Don't use flip or rotation