The inference, i.e. the application of a trained model to test images, is very slow for me. Are there any ways to speed it up?

Sorry, I don't really understand this loss function. It just seems to be putting a weight on Lb. And W is a H×W×1 map, Lb is a value, how do they work it out between them. And how to "increase the weight of boundary region noise prediction by W"?

When i testing with dsd_01.yaml, I meet the error.

How many times have you trained to get good results and how small should the loss be?

Hello, I noticed that in your code, the training and testing sets are placed in the same folder. So, during testing, a portion of the images are randomly selected for segmentation testing? Or will the last 100 images be used for testing? If you could reply, I would greatly appreciate it

In testing DermoSegDiff for some images in googleColab I am experiencing CUDA plugin conflicts (cuDNN, cuFFT, cuBLAS). Here is the error file.
error.txt

When I make minor changes to the code based on the source code, I get exceptions to the metrics.Either part of the metric is 0, or it is 20% off the source code metric.How to solve it?

Inside of this folder the required numpy files will create inside of folder.
Does this sentence mean to convert the images and Groundtruth into numpy format data and then put them in the same folder?