High-resolution rectified gradient-based visual explanations for weakly supervised segmentation

Efficient end-to-end learning for cell segmentation with machine generated weak annotations

“…A method to train instance segmentation models using LOI supervision has been proposed, but the model performance is lower 27 . An attractive alternative approach is to rely on imagelevel annotations, instead of instance-level annotations [28][29][30][31] . For example, class-activation maps 32 can be utilized as a stand-in for segmentation masks [28][29][30] because high gradients tend to localize to regions of importance for the modeling objective.…”

“…An attractive alternative approach is to rely on imagelevel annotations, instead of instance-level annotations [28][29][30][31] . For example, class-activation maps 32 can be utilized as a stand-in for segmentation masks [28][29][30] because high gradients tend to localize to regions of importance for the modeling objective. This allows the model to be trained on simpler auxiliary tasks, such as image classification.…”