def get_latent_batch(self, images):
start = time.time()
for i, image in enumerate(images):
hax_fn = "/home/" + username(
) + "/dlrc17-gdk/ext/OnAVOS/custom_dataset/JPEGImages/480p/live_fol/00001_" + str(
i + 1).zfill(4) + ".png"
# hax_fn = "/home/" + username() + "/Documents/class_data_final/live_fol/00001_" + str(i).zfill(4) + ".png"
scipy.misc.imsave(hax_fn, image)
out_layer_name = self.config.unicode("output_embedding_layer",
"outputTriplet")
out_layer = self.test_network.tower_layers[0][out_layer_name]
assert len(out_layer.outputs) == 1
out_feature = out_layer.outputs[0]
out_feature_size = out_layer.n_features
path_name = "live_fol/"
path = self.valid_data.test_case
idx_placeholder = self.valid_data.idx_placeholder
num_images = len(images)
idx_value = [0, num_images, 0, 0]
feature_val = self.session.run(out_feature,
feed_dict={
idx_placeholder: idx_value,
path: path_name
})
end = time.time()
elapsed = end - start
# print(elapsed)
return feature_val
def __init__(self, config, subset, coord, fraction=1.0):
data_dir = "/fastwork/" + username() + "/mywork/data/PascalVOC/benchmark_RELEASE/dataset/"
super(PascalVOCInteractiveCorrectionFeed, self).__init__(config, subset, coord, data_dir, fraction,
NUM_CLASSES, VOID_LABEL, INPUT_SIZE,
use_old_label=True, use_clicks=True)
self.fraction = fraction
self.old_label_as_distance_transform = config.bool("old_label_as_distance_transform", True)
def __init__(self, config, subset, coord, fraction=1.0):
data_dir = "/fastwork/" + username() + "/mywork/data/PascalVOC/benchmark_RELEASE/dataset/"
self.use_old_label = config.bool("use_old_label", False)
super(PascalVOCInteractiveFeed, self).__init__(config, subset, coord, data_dir, fraction,
NUM_CLASSES, VOID_LABEL, INPUT_SIZE,
use_old_label=self.use_old_label, use_clicks=True)
self.fraction = fraction
def __init__(self, engine):
super(ClusteringForwarder, self).__init__(engine)
self.output_folder = "/home/" + username(
) + "/vision/clustering/data/new_forward/"
self.RCNN = engine.config.bool("forward_rcnn", False)
self.network = None
self.data = None
self.output_file_name = engine.config.str("output_file_name", "")
def segment(self, image, size, save_flag=False, threshold=0.5):
start = time.time()
orig_size = (int(image.shape[1]), int(image.shape[0]))
resized_image = scipy.misc.imresize(image, size)
# hax_fn = "/home/"+username()+"/dlrc17-gdk/ext/OnAVOS/custom_dataset/JPEGImages/480p/live/000000.jpg"
hax_fn = "/home/" + username(
) + "/Documents/Segment/OnAVOSold/custom_dataset/JPEGImages/480p/live/000000.jpg"
scipy.misc.imsave(hax_fn, resized_image)
tensor_out = self.test_network.y_softmax
tensor_out_argmax = tf.argmax(tensor_out, 3)
feed_dict = {}
# feed_dict = {self.valid_data.img_placeholder: [image]}
# feed_dict = {self.test_network.img: [image]}
prob, y_argmax = self.session.run([tensor_out, tensor_out_argmax],
feed_dict=feed_dict)
prob = prob[0, :, :, 1]
# print(prob.max(),prob.min(),prob.mean())
orig_size_prob = scipy.misc.imresize(prob, orig_size, interp='bicubic')
# print(orig_size_prob.max(), orig_size_prob.min(), orig_size_prob.mean())
mask = ((orig_size_prob > 255 * threshold) * 255).astype("uint8")
# mask = (y_argmax * 255).astype("uint8")
# mask = numpy.squeeze(mask, axis=0)
# prob = numpy.squeeze(prob[:, :, :, 1], axis=0)
# mask = numpy.fliplr(mask)
# prob = numpy.fliplr(prob)
mask = scipy.misc.imresize(mask, size)
prob = scipy.misc.imresize(prob, size)
# if save_flag:
# saver_fol = "/home/dlrc/dlrc17-gdk/gdk/imagproc/video_output/"
# # fol_num = str(len(os.listdir(saver_fol))).zfill(4)
# # if not os.path.exists(fol_num):
# # os.makedirs(fol_num)
# # dir = saver_fol+fol_num+"/"
# dir = saver_fol
# im_num = str(len(os.listdir(dir))).zfill(4)
# scipy.misc.imsave(dir+"/"+im_num+".jpg", image)
# scipy.misc.imsave(dir + "/" + im_num + ".png", mask)
end = time.time()
elapsed = end - start
return mask, prob
import glob
import tensorflow as tf
from datasets.Dataset import ImageDataset
from datasets.Util.Util import username, unique_list
from datasets.Util.Reader import load_label_default
from datasets.Util.Normalization import normalize
NUM_CLASSES = 2
VOID_LABEL = 255 # for translation augmentation
DAVIS_DEFAULT_PATH = "/fastwork/" + username() + "/mywork/data/DAVIS/"
# DAVIS2017_DEFAULT_PATH = "/fastwork/" + username() + "/mywork/data/DAVIS2017/"
#DAVIS2017_DEFAULT_PATH = "/home/" + username() + "/dlrc17-gdk/ext/OnAVOS/custom_dataset/"
DAVIS2017_DEFAULT_PATH = "/home/" + username(
) + "/Documents/Segment/OnAVOSold/custom_dataset/"
DAVIS_FLOW_DEFAULT_PATH = "/fastwork/" + username(
) + "/mywork/data/DAVIS_data/"
DAVIS_LUCID_DEFAULT_PATH = "/fastwork/" + username(
) + "/mywork/data/DAVIS_data/lucid/"
DAVIS2017_LUCID_DEFAULT_PATH = "/fastwork/" + username(
) + "/mywork/data/DAVIS2017_data/lucid/"
DAVIS_IMAGE_SIZE = (480, 854)
DAVIS2017_IMAGE_SIZE = (480, None)
def read_image_and_annotation_list(fn, data_dir):
imgs = []
ans = []
with open(fn) as f:
for l in f:
sp = l.split()
an = data_dir + sp[1]
#!/usr/bin/env python
import sys
import os
from datasets.Util.Util import username
# base_file=os.getcwd()
base_file = '/home/'+username()+'/dlrc17-gdk/ext/OnAVOS'
os.chdir(base_file)
sys.path.insert(0,'./')
sys.path.insert(0,'datasets')
from Engine import Engine
from Config import Config
from Log import log
import tensorflow as tf
def init_log(config):
log_dir = config.dir("log_dir", "logs")
model = config.unicode("model")
filename = log_dir + model + ".log"
verbosity = config.int("log_verbosity", 3)
log.initialize([filename], [verbosity], [])
def main(_):
assert len(sys.argv) == 2, "usage: main.py <config>"
config_path = sys.argv[1]
assert os.path.exists(config_path), config_path
try:
config = Config(config_path)
import glob
import tensorflow as tf
from datasets.Dataset import ImageDataset
from datasets.Util.Util import username, unique_list
from datasets.Util.Reader import load_label_default
NUM_CLASSES = 2
VOID_LABEL = 255 # for translation augmentation
DAVIS_DEFAULT_PATH = "/fastwork/" + username() + "/mywork/data/DAVIS/"
DAVIS2017_DEFAULT_PATH = "/fastwork/" + username() + "/mywork/data/DAVIS2017/"
DAVIS_FLOW_DEFAULT_PATH = "/fastwork/" + username(
) + "/mywork/data/DAVIS_data/"
DAVIS_LUCID_DEFAULT_PATH = "/fastwork/" + username(
) + "/mywork/data/DAVIS_data/lucid/"
DAVIS2017_LUCID_DEFAULT_PATH = "/fastwork/" + username(
) + "/mywork/data/DAVIS2017_data/lucid/"
#DAVIS_IMAGE_SIZE = (480, None)#854)
DAVIS_IMAGE_SIZE = (None, None) #854)
DAVIS2017_IMAGE_SIZE = (480, None)
def read_image_and_annotation_list(fn, data_dir):
imgs = []
ans = []
with open(fn) as f:
for l in f:
sp = l.split()
an = data_dir + sp[1]
im = data_dir + sp[0]
imgs.append(im)
ans.append(an)
import tensorflow as tf
import numpy
import os
from datasets.Dataset import Dataset
from datasets.Util.Reader import load_image_tensorflow, load_normalized_image_tensorflow
from datasets.Util.Util import smart_shape, username
from datasets.Augmentors import parse_augmentors, apply_augmentors
from datasets.Util.Normalization import normalize
# CUHK03_DEFAULT_PATH = "/fastwork/" + username() + "/mywork/data/CUHK03/"
CUHK03_DEFAULT_PATH = "/home/" + username(
) + "/dlrc17-gdk/ext/OnAVOS/custom_dataset/JPEGImages/480p/"
DEFAULT_INPUT_SIZE = [270, 90]
CUHK03_VOID_LABEL = 255
class CUHK03Dataset(Dataset):
def __init__(self, config, subset, coord):
super(CUHK03Dataset, self).__init__(subset)
assert subset in ("train", "valid"), subset
self.ignore_classes = None
self.config = config
self.subset = subset
self.coord = coord
self.data_dir = config.unicode("data_dir", CUHK03_DEFAULT_PATH)
self.model = config.unicode("model", "")
self.train_folder = config.unicode("train_folder", "train/")
self.epoch_length = config.int("epoch_length", 1000)
self.n_classes = config.int("num_classes", None)
self.input_size = config.int_list("input_size", DEFAULT_INPUT_SIZE)
from datasets.Dataset import ImageDataset
from datasets.Util.Util import username
from datasets.Util import Reader
VOID_LABEL = 255
PASCAL_VOC_DEFAULT_PATH = "/fastwork/" + username(
) + "/mywork/data/PascalVOC/benchmark_RELEASE/dataset/"
INPUT_SIZE = (None, None)
class PascalVOCDataset(ImageDataset):
def __init__(self,
config,
subset,
coord,
label_postproc_fn,
name,
num_classes,
label_load_fn=Reader.load_label_default,
img_load_fn=Reader.load_img_default,
fraction=1.0,
ignore_classes=[]):
super(PascalVOCDataset,
self).__init__(name,
PASCAL_VOC_DEFAULT_PATH,
num_classes,
config,
subset,
coord,
INPUT_SIZE,
VOID_LABEL,
