search

Shathe / Semantic-Segmentation-Tensorflow-Eager

An example of semantic segmentation using tensorflow in eager execution.
26 stars 3 forks source link

Why are all the predicted data sets black? #11

Closed zhenyangGao closed 5 years ago

zhenyangGao commented 5 years ago

First of all, thank you for sharing. I use the original dataset to get the correct results, but the final predicted result image of the cityscapes dataset using RGB annotations is all black, and the output is black when I test with my own dataset. Do you know why? thank you very much!

Shathe commented 5 years ago

Can it be that the otput values are from the value 0 to the value n_classes? That may be the reason for seeing it black, that the values are very low

zhenyangGao commented 5 years ago

I use my own dataset, the input image is an RGB image, and the label is a binary image of 0 and 255.

  1. During training, I just want to try it, epoch=10, is the epoch I gave too small?
  2. The input image is RGB image is jpg format, the label is png format, is it caused by two different formats?
  3. There are only a few training sets and test sets, only 10.
  4. Does the channels variable in the loder function affect the binary image of the label? I am still testing, I hope to get your help! Thank you I still tested the image with the cityscapes RGB annotation, and the output is only grayed out. The previously tested dataset and epoch are relatively small. I will continue to try to increase the two and continue to try.
zhenyangGao commented 5 years ago

I solve it，thanks.

Shathe commented 5 years ago

Sorry, I had not time to reply so far. What was it?