Loss function decreasing

[Mahmood1968]

The loss function never decrease or keeps fluctuating between two numbers . I tried to used different batch size , epoch and learning rate between 0.01 and 0.001. But not that much improvement . What suggestions that you have to tune these parameters . is there any other reason other than these parameters

[hhbyyh]

I used a small part of the images as training data (around 200 images) and got the above result. Please check:

1. In the tensorboard, you can adjust the smooth rate. Move it to 0 allows you to see the raw values.
2. Usually the loss will decrease in the first a few epochs, then it's likely to keep stable.