Visualisation results on the Shanghai dataset

[Pipilong-coder]

Hello!
I would like to ask you, is that after training Shanghai dataset, when I visualise it on the test set, I found that almost every sample shows intensity overestimation in the first few frames of the prediction result, that is, it shows the [255,255,255,255] band of colours, but the real value is not there, what is the reason for this?

[DeminYu98]

hi, thanks for following our work!
From your description, I'm unable to pinpoint the problem. Could you please provide a detailed description of the training process? Based on what you've described, you might want to consider the following:

• Check if the output of mu is normal to determine if there are issues with the overall training parameters.
• Check if the model's input and ground truth have been pre-processed correctly.
• The convergence of diff_loss.