fuy34 / superpixel_fcn

[CVPR‘20] SpixelFCN: Superpixel Segmentation with Fully Convolutional Network
Other
392 stars 80 forks source link

A rectangular grid appears in the test results #38

Closed BJHYZJ closed 2 years ago

BJHYZJ commented 2 years ago

Hello author, thank you for your contribution. I retrained after modifying the model architecture, but the matrix grid appeared in the results during the test, which seemed to be initialized superpixels. I don’t know if this is normal. 206097_img_sPixel

fuy34 commented 2 years ago

Well, what is the training looks like? And could you check this image with our pre-trained model? Personally, I believe our model can perform better than this.

BJHYZJ commented 2 years ago

Yes, you are working really well, I used this image because I wanted to show you the problem with squares, I recently discovered that the initialization of a superpixel is to split the image into regular rectangles, and then with With increasing iterations, these rectangles will gradually deform into random superpixel blocks. So I want to ask you, is this normal?

BJHYZJ commented 2 years ago

image like this

fuy34 commented 2 years ago

The training looks not right to me. Hope you have solved the problem, as you closed the issue. :)