zhixuhao / unet

unet for image segmentation
MIT License
4.6k stars 2k forks source link

proplems in predict_generator #172

Open tamaraalshekhli opened 4 years ago

tamaraalshekhli commented 4 years ago

hi there, I did several changes on the model to deal with multi task problem (one image and two mask) to segment the road and their centerlines from aerial images, all the problems I faced are in the predict_generator, first thing I have 1584 samples in test set results = model.predict_generator(testGene,1584,verbose=1) the output was only two image one for the road and the other one for the centerline that's it, I didn't get the predictions to the whole 1584 images, the second thing is these two output images have a lot of problems, first thing the size of the predicted images instead of the target size (224x224) Ive got (224,1584) where 1584 refer to the number of test set samples , second thing the prediction is not even close to a road or centerline it gave a bunch of random lines, is there anyone has an idea for whats going on , any help will be appreciated