I am building U-Net Image Segmentation model to detect indvidual walrus in images. The problem I am facing is that the model is detecting groups of walruses as one blob rather than detecting the different indviduals. Example:
How can I get the model to detect individual walrus as in the true mask rather than detecting a blob of walrus as in the predicted mask?
TensorFlow tutorial I’m following: Image segmentation | TensorFlow Core
Hi @Raunak_Singh,
As your model is built on U-Net Segmentation, obviously it performs semantic segmentation on the given image. If the model required to be segment the individual walrus, 'object detection models are at first preference. As these models perform instance segmentation, they can detect individually. Take a look at different object detection models here for your use case.
Alternatively, if you want to continue with U-Net segmentation, apply Post processing methods like Watershed algorithm and Distance transform methods.
Thank You
