raghakot / ultrasound-nerve-segmentation Goto Github PK

Getting that error at line 65 of train.py:
model.fit_generator(train_generator, validation_data=val_generator, nb_val_samples=X_val.shape[0] * 2,
samples_per_epoch=X_train.shape[0], nb_epoch=nb_epoch, verbose=1,
callbacks=[model_checkpoint, reduce_lr, tb], max_q_size=10000)

At callbacks of above line.

I read documentation of fit_generation at keras official website and found that it works parllely with CPU and GPU. But I don't have GPU.
So, is this error is coming because of absence of GPU?

Thanks...

Hi Raghavendra,

Thanks for sharing the code. I have a question regarding generator.py. Looks like CustomDataGenerator(Iterator): only contains two methods: init and next. You mentioned that this is modified Modified keras ImageDataGenerator, which seems to me includes a lot of other methods.

Or in specific, I am just trying to understand how does your CustomDataGenerator(Iterator) work?. Thanks.

Besides, in train.py, the train_generator and val_generator are setup in different manner, may I know the difference?

train_generator = CustomDataGenerator(X_train, y_train, transform, batch_size)

val_generator = CustomDataGenerator(X_val, y_val, lambda x, y: transform(x, y, augment=False), batch_size)

I am getting following exception at line 61 in model.py :

Exception: "concat" mode can only merge layers with matching output shapes except for the concat axis. Layer shapes: [(None, 2, 20, 512), (None, 1, 20, 256)]

I am new to keras.

Thanks...

Hi,

I changed the dimension ordering of keras to 'theano' with backend 'tensorflow', dropped all images in "input" folder and then ran your code. Code is executing without any error but I am not getting any pixel value (mask) for test set.

Even I ran your code after commenting line {65, 66} of file submission.py:
#if has_masks[i, 0] < 0.5:
#masks[i, 0] *= 0.

But still there is no pixel value (mask) for any image of test set. Please help me..

Thanks...