search

WZH0120 / SAM2-UNet

SAM2-UNet: Segment Anything 2 Makes Strong Encoder for Natural and Medical Image Segmentation
Apache License 2.0
116 stars 15 forks source link

Salinet Object Detection #23

Open anilsathyan7 opened 5 days ago

anilsathyan7 commented 5 days ago

How is it different from rgb salient object detection models like birefnet, inspyrenet: https://paperswithcode.com/sota/salient-object-detection-on-dut-omron. Do these models have better accuracy than SAM2-UNet for general purpose usage?

xiongxyowo commented 5 days ago

Hi, BiRefNet and InSPyReNet use higher resolution input images (e.g., 1024×1024) and more advanced decoder designs, so they may achieve more accurate results. Theoretically, the performance of SAM2-UNet can be further improved if it is directly extended to high resolution as well. However, since the standard U-Net decoder is not designed to handle high-resolution images, this improvement may be very limited. You can try to modify the SAM2-UNet decoder to get better results.