Why is the shape of the loss curve still the same when the dataset is changed.Also, changing the batch size has no effect

[zhang197]

Thank you for your contribution!
I just changed the images in the dataset. Put the sign language's picture into openpose to get the skeleton map, and the file name remains the same as the original data set. That is to say, I used the skeleton diagram as the input image. But the shape of the loss curve is very similar, and the difference in numerical value is very small.
I don't think that should happen.Can anyone tell me what this is about.Thank you.
Moreover, the batch size in the hyper parameter setting does not seem to have been used in any Python file. I think this indicates that it is impossible to read the batch size data that I set