previtus

# numpy save and load:
np.save(open('labels_from_segments.npy', 'w'), labels_from_segments)
np.save(open('labels_from_model.npy', 'w'), labels_from_model)

labels_from_segments = np.load(open('labels_from_segments.npy'))
labels_from_model = np.load(open('labels_from_model.npy'))

previtus

def saveHistory(history_dict, filename):
    to_be_saved = data = {'S': history_dict}
    np.save(open(filename, 'w'), to_be_saved)

def loadHistory(filename):
    loaded = np.load(open(filename))
    return loaded[()]['S']

hi = model.fit(...)
saveHistory(hi.history, 'tmp_saved_history.npy')

previtus

# -*- coding: utf-8 -*-

import numpy as np
import copy

from keras.layers import Input, Dense, Conv2D, MaxPooling2D, AveragePooling2D, ZeroPadding2D, Flatten, Activation, add
from keras.optimizers import SGD
from keras.layers.normalization import BatchNormalization
from keras.models import Model
from keras import initializers

previtus

# use inkscape to convert all *.svg into *.png on Windows
for /f "tokens=* delims=\n" %i in ('dir /b *.svg') do "C:\Program Files (x86)\Inkscape\inkscape.exe" --without-gui --file="%i" --export-png="%i.png" --export-background=white --export-dpi=300

# print the names
for /f "tokens=* delims=\n" %i in ('dir /b *.svg') do echo %i

previtus

zip -r FOLDER.zip FOLDER/* -x *.h5

zip all files exluding .h5 files

previtus

### Inside Segment object
    def getGoogleViewUrls(self, PIXELS_X, PIXELS_Y, minimal_length):
        urls = []
        filenames = []

        min_allowed_distance = minimal_length

        d = 1000*distance_between_two_points(self.Start, self.End)
        number_of_splits = int(max((floor(d / min_allowed_distance)),1.0))

previtus

for file in *.jpg; do convert -crop 640x615+0+0 $file $file; done

previtus

# tensorflow inside keras metric, with debugging tf variables

def grouped_mse(k=3):
    def f(y_true, y_pred):
        group_by = tf.constant(k)
        real_size = tf.size(y_pred)

        remainder = tf.truncatemod(real_size, group_by)
        remainder = K.print_tensor(remainder, message="remainder is: ")

previtus

a = "/home/ekmek/saliency_tools/models/TEST1.npy"
b = "/home/ekmek/saliency_tools/models/TEST2.npy"

aval = np.load(a)
bval = np.load(b)

# lengths
print len(aval)
print len(bval)

previtus

def short_summary(model):
    from keras import backend as K
    for layer in model.layers:
        trainable_count = int( np.sum([K.count_params(p) for p in set(layer.trainable_weights)]))
        non_trainable_count = int( np.sum([K.count_params(p) for p in set(layer.non_trainable_weights)]))
        if trainable_count == 0 and non_trainable_count == 0:
            print '{:<10}[{:<10}]: {:<20} => {:<20}'.format(layer.name, layer.__class__.__name__, layer.input_shape,layer.output_shape)
        else:
            print '{:<10}[{:<10}]: {:<20} => {:<20}, with {} trainable + {} nontrainable'.format(layer.name, layer.__class__.__name__, layer.input_shape, layer.output_shape, trainable_count, non_trainable_count)