Closed jorgelerre closed 1 month ago
@jorgelerre Is this model params trained form your datatset or mine?
If you want to get good results, you must train your model based your dataset. Mine model params is trained based on sentinel-2 image.
@SonwYang I'm using your weights to run the model. However, I took the image from your drive (in the polygonize zip you have some patches, the one I tried was temp_pic/0_0.png), so I suppose it was extracted from Sentinel-2 too.
No, this image is not sentinel-2. Actually, this is china statellite image.
You can use SAM(segment anything model) to get zero-shot result. My method is not zero-shot.
Hello, SonwYang.
I'm trying to use your model with the following image:![image](https://github.com/user-attachments/assets/cb845586-d5b9-4909-ac4d-173f70e41d83)
However, the output for the edge detection is this:![image](https://github.com/user-attachments/assets/1331e7af-1456-49ff-9d00-9e81a4e6eaa4)
I saw that the prediction is generated by merging some modifications of the image (flipping, using mirror padding, etc.), done in input_and_output function. I will try to fix it by myself, but it feels pretty strange that it worked for you properly. May I be doing something wrong?