AutoSAM: Adapting SAM to Medical Images by Overloading the Prompt Encoder
This work addresses the challenge of applying SAM to medical image segmentation, an incremental improvement for the medical imaging domain.
The authors tackled the problem of adapting the Segment Anything Model (SAM) to medical images, which it fails on due to out-of-distribution issues, by replacing its conditioning with an encoder that operates on the input image, achieving state-of-the-art results on multiple medical benchmarks without fine-tuning SAM.
The recently introduced Segment Anything Model (SAM) combines a clever architecture and large quantities of training data to obtain remarkable image segmentation capabilities. However, it fails to reproduce such results for Out-Of-Distribution (OOD) domains such as medical images. Moreover, while SAM is conditioned on either a mask or a set of points, it may be desirable to have a fully automatic solution. In this work, we replace SAM's conditioning with an encoder that operates on the same input image. By adding this encoder and without further fine-tuning SAM, we obtain state-of-the-art results on multiple medical images and video benchmarks. This new encoder is trained via gradients provided by a frozen SAM. For inspecting the knowledge within it, and providing a lightweight segmentation solution, we also learn to decode it into a mask by a shallow deconvolution network.