search

GeorgeSeif / Semantic-Segmentation-Suite

Semantic Segmentation Suite in TensorFlow. Implement, train, and test new Semantic Segmentation models easily!
2.51k stars 879 forks source link

CamVid database labels #6

Closed 1451154 closed 6 years ago

1451154 commented 6 years ago

Thanks for your code！CamVid database's label is color image(RGB)，but your train_labels、val_labels and test_labels are gray image。Can you tell me why you want to convert these annotated color images into grayscale images and how to convert them? This is your train_labels: 0001tp_006690 This is CamVid' label: 0001tp_006690_l

GeorgeSeif commented 6 years ago

I'm actually using an edited version of CamVid which is commonly used in research papers. With this different version, each pixel has an integer value of its class from 0 to 11 (CamVid had 12 classes at the time).

Though it does seem you found a flaw I shall fix! As the "one_hot_it" function in "helpers.py" won't currently work with those colour labels. I'll try and fix this ASAP!

GeorgeSeif commented 6 years ago

Updated! It now works with the colour labels and thus the original CamVid dataset!

chenwydj commented 5 years ago

I'm actually using an edited version of CamVid which is commonly used in research papers. With this different version, each pixel has an integer value of its class from 0 to 11 (CamVid had 12 classes at the time).

Though it does seem you found a flaw I shall fix! As the "one_hot_it" function in "helpers.py" won't currently work with those colour labels. I'll try and fix this ASAP!

Hi George! I found in the "edited version of CamVid" the image and label has size of (320, 480) which is different from the origin size (720, 960). How did you solve this issue when you were using the "edited version of CamVid" for training and evaluation?

Thank you!