lyakaap / kaggle-carvana-3rd-place-solution Goto Github PK

Great stuff here. Any chance you could release the pretrained weights?

my doubt is regarding the implementation loss, which is a combination of binary cross entropy and dice loss(F score), what is the intuition behind combining both, BCE is loss and other is F score, please could you provide a reason or any reference.

I am new to Keras. After learning some tutorial on the internet, I tried to train. But always got a problem.

my training code is:

`train_datagen = image.ImageDataGenerator(
rotation_range=40,
width_shift_range=0.2,
height_shift_range=0.2,
rescale=1./255,
shear_range=0.2,
zoom_range=0.2,
horizontal_flip=True,
fill_mode='nearest')

val_datagen = image.ImageDataGenerator(rescale=1./255)

train_generator = train_datagen.flow_from_directory('../images/train/cat/',
target_size=(1920,1280),
batch_size=32,
class_mode='binary')

val_generator = val_datagen.flow_from_directory('../images/val/cat/',
target_size=(1920,1280),
batch_size=8,
class_mode='binary')

model.fit_generator(
train_generator,
steps_per_epoch=11250//32,
epochs=10,
validation_data=val_generator,
validation_steps=144)

model.save_weights('../models/pet.h5')`

and the error is:

`Traceback (most recent call last):
File "", line 1, in
File "/usr/local/lib/python2.7/dist-packages/spyder/utils/site/sitecustomize.py", line 880, in runfile
execfile(filename, namespace)

File "/usr/local/lib/python2.7/dist-packages/spyder/utils/site/sitecustomize.py", line 94, in execfile
builtins.execfile(filename, *where)

File "/home/hans/Kaggle/Fine-Grained/doc/fine-grained.py", line 98, in
validation_steps=144)

File "/usr/local/lib/python2.7/dist-packages/keras/legacy/interfaces.py", line 87, in wrapper
return func(*args, **kwargs)

File "/usr/local/lib/python2.7/dist-packages/keras/engine/training.py", line 2042, in fit_generator
class_weight=class_weight)

File "/usr/local/lib/python2.7/dist-packages/keras/engine/training.py", line 1762, in train_on_batch
outputs = self.train_function(ins)

File "/usr/local/lib/python2.7/dist-packages/keras/backend/tensorflow_backend.py", line 2273, in call
**self.session_kwargs)

File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 778, in run
run_metadata_ptr)

File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 961, in _run
% (np_val.shape, subfeed_t.name, str(subfeed_t.get_shape())))

ValueError: Cannot feed value of shape (32, 1) for Tensor u'conv2d_22_target:0', which has shape '(?, ?, ?, ?)'`

could you please tell me why the implement of loss function [weight_bce_dice_loss] is different from your kernel in kaggle : kernel

diff 1: weighted bce loss
[kaggle kernel]
loss = (1. - y_true) * logit_y_pred + (1. + (weight - 1.) * y_true) * \ (K.log(1. + K.exp(-K.abs(logit_y_pred))) + K.maximum(-logit_y_pred, 0.))
[github repo]
loss = weight * (logit_y_pred * (1. - y_true) + K.log(1. + K.exp(-K.abs(logit_y_pred))) + K.maximum(-logit_y_pred, 0.))

diff2: weightd bce dice loss
[kaggle kernel]
border = K.cast(K.greater(averaged_mask, 0.005), 'float32') * K.cast(K.less(averaged_mask, 0.995), 'float32')
[github repo]
weight = 5. * K.exp(-5. * K.abs(averaged_mask - 0.5))

The bottleneck function in model.py does the same for parallel and cascade mode

what does the meaning of weight w0 & w1 in loss function "weighted_bce_dice_loss"