def load_exclusion_region_generator(config_yaml):
window_dims = tuple(map(int, config_yaml['training']['svm']['window_dims']))
bak_img_dir = config_yaml['dataset']['directory']['generation']['input']['exclusion']
bbinfo_dir = config_yaml['dataset']['directory']['bbinfo']
exl_info_map = utils.load_opencv_bounding_box_info_directory(bbinfo_dir, suffix='exclusion')
modifiers = config_yaml['dataset']['modifiers']
return generate_negative_regions_with_exclusions(bak_img_dir, exl_info_map, window_dims, modifiers)
def sampleTrainingImages(image_dir, synsets, sample_size, require_bboxes=False, bbinfo_dir=None):
image_list = utils.list_images_in_directory(image_dir)
# regexString = '.*({})_.*\.jpg'.format('|'.join(synsets))
# regexString = 'n?({})(_.*)?\.(jpg|png)'.format('|'.join(synsets))
regexString = '(.*/)?({})(_.*)?\.(jpg|png)'.format('|'.join(synsets))
synsetProg = re.compile(regexString)
# Filter image file lists to match sysnet specifications:
filtered_image_list = filter(lambda x: synsetProg.match(x), image_list)
# print regexString
# if len(regexString) > 100:
# # print image_list
# print filtered_image_list
# If bounding boxes are required, filter out images without bounding boxes:
if require_bboxes:
global_info = utils.load_opencv_bounding_box_info_directory(bbinfo_dir, suffix='bbinfo')
# global_info = loadGlobalInfo(bbinfo_dir)
bbox_filter = lambda img_path: os.path.split(img_path)[1] in global_info
filtered_image_list = filter(bbox_filter, filtered_image_list)
print len(filtered_image_list)
# Truncate file lists to sample the correct number of images:
image_sample = filtered_image_list
if not sample_size is None:
image_sample = random.sample(filtered_image_list, min(sample_size, len(filtered_image_list)))
return image_sample
def generate_positive_regions(image_dir, bbinfo_dir, modifiers_config=None, window_dims=None, min_size=(48,48)):
print 'generate_positive_regions:'
all_images = utils.list_images_in_directory(image_dir)
# Create the modifier generator:
# Note: Simply generate null modifiers if no modifier config is passed.
modifier_generator = itertools.repeat(utils.RegionModifiers())
if modifiers_config:
modifier_generator = utils.RegionModifiers.random_generator_from_config_dict(modifiers_config)
# Filter out images without bounding boxes:
bbinfo_map = utils.load_opencv_bounding_box_info_directory(bbinfo_dir, suffix='bbinfo')
source_images = [img_path for img_path in all_images if not utils.info_entry_for_image(bbinfo_map, img_path) is None]
# Extract image regions:
source_regions = []
for img_path in source_images:
rects = utils.info_entry_for_image(bbinfo_map, img_path)
for rect in rects:
# Reject small samples:
if rect.w < float(min_size[0]) or rect.h < float(min_size[1]):
continue
region = utils.ImageRegion(rect, img_path)
source_regions.append(region)
print 'Found {} source regions.'.format(len(source_regions))
# Generate an infinite list of samples:
while True:
# Randomise the order of source images:
# Note: Don't randomise if no modifiers are used.
if modifiers_config:
random.shuffle(source_regions)
# For each source region:
for reg in source_regions:
mod = modifier_generator.next()
# Enlarge to correct aspect ratio:
new_rect = reg.rect
if window_dims:
aspect = window_dims[0] / float(window_dims[1])
new_rect = new_rect.enlarge_to_aspect(aspect)
imsize = utils.get_image_dimensions(reg.fname)
if not new_rect.lies_within_frame(imsize):
# print 'bad', new_rect
continue
# else:
# print new_rect
# Assign a random modifier, and yield the region:
new_reg = utils.ImageRegion(new_rect, reg.fname, mod)
yield new_reg
# If we're not using modifiers, only use each region once:
if not modifiers_config:
break
def view_positive_samples(classifier_yaml, output_dir):
bbinfo_dir = classifier_yaml['dataset']['directory']['bbinfo']
global_info = utils.load_opencv_bounding_box_info_directory(
bbinfo_dir, suffix='bbinfo')
# global_info = loadGlobalInfo(bbinfo_dir)
positive_dir = classifier_yaml['dataset']['directory']['positive']
# for img_path in glob.glob("{}/*_*.jpg".format(positive_dir)):
# for img_path in utils.list_images_in_directory(positive_dir):
pos_samples = sampleTrainingImages(positive_dir, ['.*'],
None,
require_bboxes=True,
bbinfo_dir=bbinfo_dir)
print 'Selected {} positive samples.'.format(len(pos_samples))
for img_path in pos_samples:
img = cv2.imread(img_path)
key = img_path.split('/')[-1]
rects_str = global_info[key]
rects = utils.rectangles_from_cache_string(rects_str)
print img_path, len(rects)
for rect in rects:
lw = max(2, min(img.shape[0], img.shape[1]) / 150)
cvDrawRectangle(img, rect, (255, 0, 0), lw)
# cv2.rectangle(img,(x,y),(x+w,y+h),(255,255,255),lw+2)
# cv2.rectangle(img,(x,y),(x+w,y+h),(255,0,0),lw)
# aspectRect = gm.extendBoundingBox(rect, 83/64.0)
# cvDrawRectangle(img, aspectRect, (0,255,0),2)
#
# paddedRect = gm.padBoundingBox(aspectRect, (0.1, 0.1))
# cvDrawRectangle(img, paddedRect, (0,0,255),2)
h, w = img.shape[:2]
max_dim = 1200.0
# if w > max_dim + 50 or h > max_dim + 50:
if abs(w - max_dim) > 50 or abs(h - max_dim) > 50:
sx = max_dim / w
sy = max_dim / h
current_scale = min(sx, sy)
img = cv2.resize(img,
dsize=None,
fx=current_scale,
fy=current_scale)
cv2.imshow('img', img)
cv2.waitKey(0)
cv2.destroyAllWindows()
def checkBoundingBoxes(img_paths, bbinfo_dir):
global_info = utils.load_opencv_bounding_box_info_directory(
bbinfo_dir, suffix='bbinfo')
# global_info = loadGlobalInfo(bbinfo_dir)
error_occurred = False
bad_images = []
warn_images = []
for img_path in img_paths:
imsize = None
try:
im = Image.open(img_path)
imsize = im.size
im.close()
except Exception as e:
print "Error opening image {}: {}".format(img_path, e)
error_occurred = True
continue
# Get bounding box:
rects = utils.info_entry_for_image(global_info, img_path)
for rect in rects:
if not rect.touches_frame_edge(imsize):
warn_images.append((img_path, rect, imsize))
if not rect.lies_within_frame(imsize):
bad_images.append((img_path, rect, imsize))
if warn_images:
print """
WARNING: Some bounding boxes touch the image edges.
An older version of this script defined bounding boxes differently
in which case the case where rect.x + rect.w == img_width would
indicate a bounding box extending outside of the image dimensions.
The newer KITTI scripts define bounding box size as the true pixel
size, making this test indicate a box touching the image edge, which
is fine.
If the OpenCV scripts fail, this may be the reason.
(Uncomment the following line to see the suspicious boxes)
"""
# pprint(warn_images)
if error_occurred or bad_images:
print 'EXITING DUE TO INVALID BOUNDING BOXES:'
pprint(bad_images)
sys.exit(1)
assert (False)
def checkBoundingBoxes(img_paths, bbinfo_dir):
global_info = utils.load_opencv_bounding_box_info_directory(bbinfo_dir, suffix='bbinfo')
# global_info = loadGlobalInfo(bbinfo_dir)
error_occurred = False
bad_images = []
warn_images = []
for img_path in img_paths:
imsize = None
try:
im = Image.open(img_path)
imsize = im.size
im.close()
except Exception as e:
print "Error opening image {}: {}".format(img_path, e)
error_occurred = True
continue
# Get bounding box:
rects = utils.info_entry_for_image(global_info, img_path)
for rect in rects:
if not rect.touches_frame_edge(imsize):
warn_images.append((img_path, rect, imsize))
if not rect.lies_within_frame(imsize):
bad_images.append((img_path, rect, imsize))
if warn_images:
print """
WARNING: Some bounding boxes touch the image edges.
An older version of this script defined bounding boxes differently
in which case the case where rect.x + rect.w == img_width would
indicate a bounding box extending outside of the image dimensions.
The newer KITTI scripts define bounding box size as the true pixel
size, making this test indicate a box touching the image edge, which
is fine.
If the OpenCV scripts fail, this may be the reason.
(Uncomment the following line to see the suspicious boxes)
"""
# pprint(warn_images)
if error_occurred or bad_images:
print 'EXITING DUE TO INVALID BOUNDING BOXES:'
pprint(bad_images)
sys.exit(1)
assert(False)
def view_positive_samples(classifier_yaml, output_dir):
bbinfo_dir = classifier_yaml['dataset']['directory']['bbinfo']
global_info = utils.load_opencv_bounding_box_info_directory(bbinfo_dir, suffix='bbinfo')
# global_info = loadGlobalInfo(bbinfo_dir)
positive_dir = classifier_yaml['dataset']['directory']['positive']
# for img_path in glob.glob("{}/*_*.jpg".format(positive_dir)):
# for img_path in utils.list_images_in_directory(positive_dir):
pos_samples = sampleTrainingImages(positive_dir, ['.*'], None, require_bboxes=True, bbinfo_dir=bbinfo_dir)
print 'Selected {} positive samples.'.format(len(pos_samples))
for img_path in pos_samples:
img = cv2.imread(img_path)
key = img_path.split('/')[-1]
rects_str = global_info[key]
rects = utils.rectangles_from_cache_string(rects_str)
print img_path, len(rects)
for rect in rects:
lw = max(2, min(img.shape[0], img.shape[1]) / 150)
cvDrawRectangle(img, rect, (255,0,0), lw)
# cv2.rectangle(img,(x,y),(x+w,y+h),(255,255,255),lw+2)
# cv2.rectangle(img,(x,y),(x+w,y+h),(255,0,0),lw)
# aspectRect = gm.extendBoundingBox(rect, 83/64.0)
# cvDrawRectangle(img, aspectRect, (0,255,0),2)
#
# paddedRect = gm.padBoundingBox(aspectRect, (0.1, 0.1))
# cvDrawRectangle(img, paddedRect, (0,0,255),2)
h, w = img.shape[:2]
max_dim = 1200.0
# if w > max_dim + 50 or h > max_dim + 50:
if abs(w - max_dim) > 50 or abs(h - max_dim) > 50:
sx = max_dim / w
sy = max_dim / h
current_scale = min(sx, sy)
img = cv2.resize(img, dsize=None, fx=current_scale, fy=current_scale)
cv2.imshow('img', img)
cv2.waitKey(0)
cv2.destroyAllWindows()
def sampleTrainingImages(image_dir,
synsets,
sample_size,
require_bboxes=False,
bbinfo_dir=None):
image_list = utils.list_images_in_directory(image_dir)
# regexString = '.*({})_.*\.jpg'.format('|'.join(synsets))
# regexString = 'n?({})(_.*)?\.(jpg|png)'.format('|'.join(synsets))
regexString = '(.*/)?({})(_.*)?\.(jpg|png)'.format('|'.join(synsets))
synsetProg = re.compile(regexString)
# Filter image file lists to match sysnet specifications:
filtered_image_list = filter(lambda x: synsetProg.match(x), image_list)
# print regexString
# if len(regexString) > 100:
# # print image_list
# print filtered_image_list
# If bounding boxes are required, filter out images without bounding boxes:
if require_bboxes:
global_info = utils.load_opencv_bounding_box_info_directory(
bbinfo_dir, suffix='bbinfo')
# global_info = loadGlobalInfo(bbinfo_dir)
bbox_filter = lambda img_path: os.path.split(img_path)[1
] in global_info
filtered_image_list = filter(bbox_filter, filtered_image_list)
print len(filtered_image_list)
# Truncate file lists to sample the correct number of images:
image_sample = filtered_image_list
if not sample_size is None:
image_sample = random.sample(
filtered_image_list, min(sample_size, len(filtered_image_list)))
return image_sample
def create_or_load_descriptors(classifier_yaml, hog, window_dims):
pos_img_dir = classifier_yaml["dataset"]["directory"]["positive"]
pos_num = int(classifier_yaml["training"]["svm"]["pos_num"])
bbinfo_dir = classifier_yaml["dataset"]["directory"]["bbinfo"]
bak_img_dir = classifier_yaml["dataset"]["directory"]["background"]
neg_num = int(classifier_yaml["training"]["svm"]["neg_num"])
# pos_descriptor_file = 'output/pos_descriptors.pickle'
# neg_descriptor_file = 'output/neg_descriptors.pickle'
# if not (os.path.isfile(pos_descriptor_file) and os.path.isfile(neg_descriptor_file)):
store = DataStore()
print "db_name:", store.db_name_for_hog(hog)
print "hog_name:", utils.name_from_hog_descriptor(hog)
# if not store.has_region_descriptors_for_hog(hog):
# Preprocess samples:
# - generate/select sample regions
# - compute and descriptors for all samples
# - save the regions and descriptors to a file
curr_pos_num = store.num_region_descriptors(hog, 1)
req_pos_num = pos_num if curr_pos_num is None else pos_num - curr_pos_num
print "pos_num:", pos_num
print "curr_pos_num:", curr_pos_num
print "req_pos_num:", req_pos_num
if req_pos_num > 0:
# if not os.path.isfile(pos_descriptor_file):
# pos_reg_generator = generate_positive_regions(pos_img_dir, bbinfo_dir, classifier_yaml['dataset']['modifiers'], 0.5*np.array(window_dims))
pos_reg_generator = generate_samples.generate_positive_regions(
pos_img_dir, bbinfo_dir, classifier_yaml["dataset"]["modifiers"], window_dims
)
# view_image_regions(pos_reg_generator, window_dims, display_scale=3)
pos_regions = list(itertools.islice(pos_reg_generator, 0, req_pos_num))
pos_reg_descriptors = compute_hog_descriptors(hog, pos_regions, window_dims, 1)
# save_region_descriptors(hog, pos_regions, pos_descriptors, 'pos_descriptors')
store.save_region_descriptors(pos_reg_descriptors, hog)
curr_neg_num = store.num_region_descriptors(hog, -1)
req_neg_num = neg_num if curr_neg_num is None else neg_num - curr_neg_num
print "neg_num:", neg_num
print "curr_neg_num:", curr_neg_num
print "req_neg_num:", req_neg_num
if req_neg_num > 0:
# if not os.path.isfile(neg_descriptor_file):
# neg_regions = generate_negative_regions(bak_img_dir, req_neg_num, window_dims)
exl_info_map = utils.load_opencv_bounding_box_info_directory(bbinfo_dir, suffix="exclusion")
neg_reg_generator = generate_samples.generate_negative_regions_with_exclusions(
bak_img_dir, exl_info_map, window_dims, classifier_yaml["dataset"]["modifiers"]
)
neg_regions = list(itertools.islice(neg_reg_generator, 0, req_neg_num))
neg_reg_descriptors = compute_hog_descriptors(hog, neg_regions, window_dims, -1)
# save_region_descriptors(hog, neg_regions, neg_descriptors, 'neg_descriptors')
store.save_region_descriptors(neg_reg_descriptors, hog)
# else:
# Load all descriptors and the hog object used to generate them:
# hog, reg_descriptors = store.load_region_descriptors_for_hog(hog)
hog, pos_reg_descriptors = store.load_region_descriptors_for_hog(hog, pos_num, 1)
hog, neg_reg_descriptors = store.load_region_descriptors_for_hog(hog, neg_num, -1)
# hog, reg_descriptors = store.load_region_descriptors('hog_Llht0,2_n64_gcFalse_hnt0_bs8x8_cs8x8_ws4,0_da1')
# hog, pos_regions, pos_descriptors = load_region_descriptors(pos_descriptor_file)
# hog, neg_regions, neg_descriptors = load_region_descriptors(neg_descriptor_file)
reg_descriptors = list(pos_reg_descriptors) + list(neg_reg_descriptors)
return reg_descriptors, hog
def create_or_load_descriptors(classifier_yaml, hog, window_dims):
pos_img_dir = classifier_yaml['dataset']['directory']['positive']
pos_num = int(classifier_yaml['training']['svm']['pos_num'])
bbinfo_dir = classifier_yaml['dataset']['directory']['bbinfo']
bak_img_dir = classifier_yaml['dataset']['directory']['background']
neg_num = int(classifier_yaml['training']['svm']['neg_num'])
# pos_descriptor_file = 'output/pos_descriptors.pickle'
# neg_descriptor_file = 'output/neg_descriptors.pickle'
# if not (os.path.isfile(pos_descriptor_file) and os.path.isfile(neg_descriptor_file)):
store = DataStore()
print 'db_name:', store.db_name_for_hog(hog)
print 'hog_name:', utils.name_from_hog_descriptor(hog)
# if not store.has_region_descriptors_for_hog(hog):
# Preprocess samples:
# - generate/select sample regions
# - compute and descriptors for all samples
# - save the regions and descriptors to a file
curr_pos_num = store.num_region_descriptors(hog, 1)
req_pos_num = pos_num if curr_pos_num is None else pos_num - curr_pos_num
print 'pos_num:', pos_num
print 'curr_pos_num:', curr_pos_num
print 'req_pos_num:', req_pos_num
if req_pos_num > 0:
# if not os.path.isfile(pos_descriptor_file):
# pos_reg_generator = generate_positive_regions(pos_img_dir, bbinfo_dir, classifier_yaml['dataset']['modifiers'], 0.5*np.array(window_dims))
pos_reg_generator = generate_samples.generate_positive_regions(
pos_img_dir, bbinfo_dir, classifier_yaml['dataset']['modifiers'],
window_dims)
# view_image_regions(pos_reg_generator, window_dims, display_scale=3)
pos_regions = list(itertools.islice(pos_reg_generator, 0, req_pos_num))
pos_reg_descriptors = compute_hog_descriptors(hog, pos_regions,
window_dims, 1)
# save_region_descriptors(hog, pos_regions, pos_descriptors, 'pos_descriptors')
store.save_region_descriptors(pos_reg_descriptors, hog)
curr_neg_num = store.num_region_descriptors(hog, -1)
req_neg_num = neg_num if curr_neg_num is None else neg_num - curr_neg_num
print 'neg_num:', neg_num
print 'curr_neg_num:', curr_neg_num
print 'req_neg_num:', req_neg_num
if req_neg_num > 0:
# if not os.path.isfile(neg_descriptor_file):
# neg_regions = generate_negative_regions(bak_img_dir, req_neg_num, window_dims)
exl_info_map = utils.load_opencv_bounding_box_info_directory(
bbinfo_dir, suffix='exclusion')
neg_reg_generator = generate_samples.generate_negative_regions_with_exclusions(
bak_img_dir, exl_info_map, window_dims,
classifier_yaml['dataset']['modifiers'])
neg_regions = list(itertools.islice(neg_reg_generator, 0, req_neg_num))
neg_reg_descriptors = compute_hog_descriptors(hog, neg_regions,
window_dims, -1)
# save_region_descriptors(hog, neg_regions, neg_descriptors, 'neg_descriptors')
store.save_region_descriptors(neg_reg_descriptors, hog)
# else:
# Load all descriptors and the hog object used to generate them:
# hog, reg_descriptors = store.load_region_descriptors_for_hog(hog)
hog, pos_reg_descriptors = store.load_region_descriptors_for_hog(
hog, pos_num, 1)
hog, neg_reg_descriptors = store.load_region_descriptors_for_hog(
hog, neg_num, -1)
# hog, reg_descriptors = store.load_region_descriptors('hog_Llht0,2_n64_gcFalse_hnt0_bs8x8_cs8x8_ws4,0_da1')
# hog, pos_regions, pos_descriptors = load_region_descriptors(pos_descriptor_file)
# hog, neg_regions, neg_descriptors = load_region_descriptors(neg_descriptor_file)
reg_descriptors = list(pos_reg_descriptors) + list(neg_reg_descriptors)
return reg_descriptors, hog
# raise ValueError('Not enough images!')
raise TooFewImagesError(presentCounts, requiredCounts)
# Create the output directory:
if not os.path.isdir(output_dir):
print '## Creating output directory: {}'.format(output_dir)
os.makedirs(output_dir)
else:
print '## Using existing output directory: {}'.format(output_dir)
# Print names of images with invalid bounding boxes:
checkBoundingBoxes(pos_image_files, bbinfo_dir)
# Load the global info file with bounding boxes for all positive images:
# global_info_fname = 'info.dat'
global_info = utils.load_opencv_bounding_box_info_directory(bbinfo_dir, suffix='bbinfo')
pos_info_fname = '{}/positive.txt'.format(output_dir)
neg_info_fname = '{}/negative.txt'.format(output_dir)
# Note: image paths in the data file have to be relative to the file itself.
abs_output_dir = os.path.abspath(output_dir)
def write_img(img_path, dat_file, write_bbox=True):
rel_img_path = None
if os.path.isabs(img_path):
rel_img_path = os.path.relpath(img_path, abs_output_dir)
else:
rel_img_path = os.path.relpath(img_path, output_dir)
dat_line = rel_img_path
# raise ValueError('Not enough images!')
raise TooFewImagesError(presentCounts, requiredCounts)
# Create the output directory:
if not os.path.isdir(output_dir):
print '## Creating output directory: {}'.format(output_dir)
os.makedirs(output_dir)
else:
print '## Using existing output directory: {}'.format(output_dir)
# Print names of images with invalid bounding boxes:
checkBoundingBoxes(pos_image_files, bbinfo_dir)
# Load the global info file with bounding boxes for all positive images:
# global_info_fname = 'info.dat'
global_info = utils.load_opencv_bounding_box_info_directory(
bbinfo_dir, suffix='bbinfo')
pos_info_fname = '{}/positive.txt'.format(output_dir)
neg_info_fname = '{}/negative.txt'.format(output_dir)
# Note: image paths in the data file have to be relative to the file itself.
abs_output_dir = os.path.abspath(output_dir)
def write_img(img_path, dat_file, write_bbox=True):
rel_img_path = None
if os.path.isabs(img_path):
rel_img_path = os.path.relpath(img_path, abs_output_dir)
else:
rel_img_path = os.path.relpath(img_path, output_dir)
dat_line = rel_img_path
