kukuruza

kukuruza / labels_gtav_to_20classes.py

Last active July 16, 2018 10:46

Convert GTA5 labels to github.com/mil-tokyo/MCD_DA compatible format (it will have 20 classes)

	'''
	For each colored label map that GTAV provides,
	create a grayscale label map for 20 classes with pixels values in [0, 19].
	'''

	import os, os.path as op
	import numpy as np
	import cv2
	from glob import glob
	from progressbar import ProgressBar

kukuruza / labels_cityscrapes_to_20classes.py

Last active May 8, 2018 19:45

Make Cityscrapes labels compatible with github.com/mil-tokyo/MCD_DA (turn 33 classes into 20 classes).

	'''
	For each gtFine_labelIds file of Cityscrapes dataset,
	create a grayscale label map synthetic<->real task
	with 20 classes with pixels values in [0, 19],
	where #19 is the background.
	The classes are compatible with github.com/mil-tokyo/MCD_DA.
	'''

	import os, os.path as op
	import numpy as np

kukuruza / update_watcher.py

Created October 4, 2017 15:03

A simple utility class to wait until a file in the filesystem has changed.

	# Usage:
	# watcher = UpdateWatcher(changing_file_path)
	# while True:
	# watcher.wait_for_update()
	# # Do my stuff.

	import os, os.path as op
	import time
	import logging

kukuruza / tf-generate-embeddings.py

Last active May 19, 2017 21:52

Example of generating embeddings for images from a saved model (tensorflow)

	# imports and other stuff


	# example constants and instances of classes
	data = Data() # some data provider
	model = Model() # some model
	batch_size = 16
	pretrained_model_path = 'model/my_model'
	log_dir = 'logs'

kukuruza / cifar10_vis_excitations.py

Last active January 30, 2023 12:56

Visualize deeper layers in Tensorflow by displaying images which gain the highest response from neurons. Written for cifar10 model.

	# Copyright 2015 Google Inc. All Rights Reserved.
	#
	# Licensed under the Apache License, Version 2.0 (the "License");
	# you may not use this file except in compliance with the License.
	# You may obtain a copy of the License at
	#
	# http://www.apache.org/licenses/LICENSE-2.0
	#
	# Unless required by applicable law or agreed to in writing, software
	# distributed under the License is distributed on an "AS IS" BASIS,

kukuruza / gist_cifar10_train.py

Last active November 4, 2024 17:36

Tensorflow: visualize convolutional filters (conv1) in Cifar10 model

	from math import sqrt

	def put_kernels_on_grid (kernel, pad = 1):

	'''Visualize conv. filters as an image (mostly for the 1st layer).
	Arranges filters into a grid, with some paddings between adjacent filters.

	Args:
	kernel: tensor of shape [Y, X, NumChannels, NumKernels]
	pad: number of black pixels around each filter (between them)

kukuruza / cifar10_train_and_eval.py

Created March 5, 2016 23:42

tensorflow: train and evaluate Cifar10 model during the same run. Train and test data are evaluated and sent to tensorboard.

	# !!! Note for cifar10_train_and_eval.py !!!
	#
	# 1. Put this file into tensorflow/models/image/cifar10 directory.
	# 2. For this file to work, you need to comment out tf.image_summary() in
	# file tensorflow/models/image/cifar_input.py
	#


	# Copyright 2015 Google Inc. All Rights Reserved.
	#