WZH0120 / SAM2-UNet

SAM2-UNet: Segment Anything 2 Makes Strong Encoder for Natural and Medical Image Segmentation
Apache License 2.0
142 stars 16 forks source link

image size #14

Open 9Cirno9 opened 1 month ago

9Cirno9 commented 1 month ago

前辈您好,非常感谢您对这个优秀工作做出的贡献! 请问关于读入图像size = 352是怎么确定的?我如果想调大image_size的话,SAM2-UNet模型构建的地方需要做更改吗?

xiongxyowo commented 1 month ago

The input size generally depends on the compared methods. For example, the common size is 256 or 384 for the polyp segmentation task. You can also try using a larger size such as 1024. Note that higher resolutions introduce greater computational overhead and do not guarantee an increase in accuracy.

xiongxyowo commented 1 month ago

In fact, the standard U-Net Decoder is not efficient at processing high resolution images. You can try replacing the SAM2-UNet decoder with some classic efficient design, such as BiSeNet.