When I used my own dataset, I encounted an error!

kwotsin / TensorFlow-ENet

TensorFlow implementation of ENet

MIT License

257 stars 123 forks source link

When I used my own dataset, I encounted an error! #35

Open Antyaoyao opened 5 years ago

Antyaoyao commented 5 years ago

When I used my own dataset, I encounted an error! My dataset contains 3 num_classes. But after I finished training my own dataset and used my trained model to predict test_images ( including 3 classes ), I just got a 2-classes segmentations. Please, thx!

white0dew commented 5 years ago

maybe you should change the parameter in the codes。

burhr2 commented 5 years ago

hi i have an issue also trying my own, i have a two class dataset here is what i am getting. How did you train yours

File "train_enet.py", line 94, in class_weights = median_frequency_balancing() File "C:\Users\User\anaconda project directory\computer vision\Deep learning\semantic segmentation\TensorFlow-ENetV1\get_class_weights.py", line 94, in median_frequency_balancing total_frequency = sum(j) / total_pixels ZeroDivisionError: division by zero

x7hkvip commented 5 years ago

check out get_class_weights.py,the code assigned weight equals 0 for last class,maybe your back ground class is 0(first one),not the last?