training my custom dataset problem #504
Description
I've seen your comments on weak or untextured datasets in a lot of issues. I am currently ranging sea surface targets, and do not use the pre-trained model to train from 0, I found that the loss of the model at the beginning is very small (10e-6 level), but the overall loss is still showing a downward trend, please ask if this model is suitable for the use of this scenario, the following is the dataset picture and the loss picture. And I found that the output of your model is between 0.1-100, not the real depth value, if I need to output the real depth value (distance value) what should I do?
