def transform_labels(bbox_labels, sample_bbox):
proj_bbox = bbox(0, 0, 0, 0)
sample_labels = []
for i in range(len(bbox_labels)):
sample_label = []
object_bbox = bbox(bbox_labels[i][1], bbox_labels[i][2],
bbox_labels[i][3], bbox_labels[i][4])
if not meet_emit_constraint(object_bbox, sample_bbox):
continue
sample_width = sample_bbox.xmax - sample_bbox.xmin
sample_height = sample_bbox.ymax - sample_bbox.ymin
proj_bbox.xmin = (object_bbox.xmin - sample_bbox.xmin) / sample_width
proj_bbox.ymin = (object_bbox.ymin - sample_bbox.ymin) / sample_height
proj_bbox.xmax = (object_bbox.xmax - sample_bbox.xmin) / sample_width
proj_bbox.ymax = (object_bbox.ymax - sample_bbox.ymin) / sample_height
proj_bbox = clip_bbox(proj_bbox)
if bbox_area(proj_bbox) > 0:
sample_label.append(bbox_labels[i][0])
sample_label.append(float(proj_bbox.xmin))
sample_label.append(float(proj_bbox.ymin))
sample_label.append(float(proj_bbox.xmax))
sample_label.append(float(proj_bbox.ymax))
sample_label.append(bbox_labels[i][5])
sample_labels.append(sample_label)
return sample_labels
def proc_fit(name, k):
im = cv2.imread(name)
img = create_fixed_image_shape(bbox(im), out_shape,
random_fill=False, mode='fit')
new_name = name.replace(in_folder, out_folder)
cv2.imwrite(new_name, img)
if k % 10 == 0:
print "Completed %d." % k
def get_image_with_bbox(attrs):
images = parse_file(config.get('deepfashion', 'attributes_file'),
val_type=int, key_item_id=None, validate_fields=False)
attrs = parse_attr(attrs, _PREDEFINED_ATTR)
filtered = filter_items(images, attrs)
image_files = append_path(config.get('deepfashion', 'image_dir'), filtered, key='image_name')
boxes = bbox(filtered)
return image_files, boxes
def satisfy_sample_constraint(sampler, sample_bbox, bbox_labels):
if sampler.min_jaccard_overlap == 0 and sampler.max_jaccard_overlap == 0:
return True
for i in range(len(bbox_labels)):
object_bbox = bbox(bbox_labels[i][1], bbox_labels[i][2],
bbox_labels[i][3], bbox_labels[i][4])
overlap = jaccard_overlap(sample_bbox, object_bbox)
if sampler.min_jaccard_overlap != 0 and overlap < sampler.min_jaccard_overlap:
continue
if sampler.max_jaccard_overlap != 0 and overlap > sampler.max_jaccard_overlap:
continue
return True
return False
def proc(name, k):
im = cv2.imread(name)
imre = create_fixed_image_shape(bbox(im), temp_shape,
random_fill=False, mode='fit')
imgs = data_augmentation(imre, frame_size=out_shape)
for i in range(len(imgs)):
img = imgs[i]
imgs.append(img[:, ::-1, :])
for i, img in enumerate(imgs):
new_name = name.replace(in_folder, out_folder)
new_name = new_name.replace('.jpeg', '_%d.jpeg' % i)
cv2.imwrite(new_name, img)
if k % 10 == 0:
print "Completed %d." % k
def generate_sample(sampler):
scale = np.random.uniform(sampler.min_scale, sampler.max_scale)
aspect_ratio = np.random.uniform(sampler.min_aspect_ratio,
sampler.max_aspect_ratio)
aspect_ratio = max(aspect_ratio, (scale**2.0))
aspect_ratio = min(aspect_ratio, 1 / (scale**2.0))
bbox_width = scale * (aspect_ratio**0.5)
bbox_height = scale / (aspect_ratio**0.5)
xmin_bound = 1 - bbox_width
ymin_bound = 1 - bbox_height
xmin = np.random.uniform(0, xmin_bound)
ymin = np.random.uniform(0, ymin_bound)
xmax = xmin + bbox_width
ymax = ymin + bbox_height
sampled_bbox = bbox(xmin, ymin, xmax, ymax)
return sampled_bbox
def test_IDT_aux(depth_image, color_image, prev_silhouette, use_bbox=False, draw_contour=True,
threshold=0, real_time=True):
normalized = normalize_depth_image(depth_image, threshold, amplify=True)
bbox = []
if use_bbox:
color_image = np.expand_dims(color_image, axis=0)
(frames_idx, bbox) = ut.bbox(color_image)
if 0 not in frames_idx:
# no person was detected in the frame
bbox = []
else:
bbox = bbox[0]
silhouette = get_img_silhouette(normalized, prev_silhouette, bbox)
IDT = ut.getIDT(silhouette)
if draw_contour:
contour = ut.createCircleContour(IMG_WIDTH, IMG_HEIGHT)
IDT_gradient = ut.getImgGradient(IDT)
normals = ut.calc_normal(contour, normalize=False).astype(int)
IDT_BGR = draw_contour_on_IDT(IDT, contour, normals, IDT_gradient)
# convert IDT to int so that we can display it with cv.imshow
IDT_BGR = cv.convertScaleAbs(IDT_BGR)
silhouette_BGR = cv.cvtColor(silhouette, cv.COLOR_GRAY2BGR)
normalized_BGR = cv.cvtColor(normalized, cv.COLOR_GRAY2BGR)
result = np.hstack((normalized_BGR, silhouette_BGR, IDT_BGR))
else:
IDT = cv.convertScaleAbs(IDT)
result = np.hstack((normalized, silhouette, IDT))
# display result
title_result = 'result: normalized, silhouette, IDT with threshold ' + \
str(threshold)
ut.imgDisplayCV(result, title_result)
if not real_time:
cv.waitKey(0)
return silhouette
def process_img(name, crop_shape, scale):
ferr = open("out_%d.log" % os.getpid(), 'a')
sys.stdout = ferr
sys.stderr = ferr
print "%s [%d] Processing file %s" % (get_time(), os.getpid(), name)
a = cv2.imread(name)
a = bbox(scaleRadius(a,scale))
if a is None:
ferr.close()
return
b = np.zeros(a.shape)
cv2.circle(b,(a.shape[1]//2,a.shape[0]//2),int(scale*0.9),(1,1,1),-1,8,0)
aa = cv2.addWeighted(a,4,cv2.GaussianBlur(a,(0,0),scale/30),-4,128)*b+128*(1-b)
rand_im = random_crops(aa, shape=crop_shape)
if "train/" in name:
new_name = name.replace("train/", "%d_train/" % scale)
elif "validation/" in name:
new_name = name.replace("validation/", "%d_val/" % scale)
cv2.imwrite(new_name,rand_im)
ferr.close()
def expand_image(img, bbox_labels, img_width, img_height):
prob = np.random.uniform(0, 1)
if prob < train_parameters['image_distort_strategy']['expand_prob']:
expand_max_ratio = train_parameters['image_distort_strategy'][
'expand_max_ratio']
if expand_max_ratio - 1 >= 0.01:
expand_ratio = np.random.uniform(1, expand_max_ratio)
height = int(img_height * expand_ratio)
width = int(img_width * expand_ratio)
h_off = math.floor(np.random.uniform(0, height - img_height))
w_off = math.floor(np.random.uniform(0, width - img_width))
expand_bbox = bbox(-w_off / img_width, -h_off / img_height,
(width - w_off) / img_width,
(height - h_off) / img_height)
expand_img = np.uint8(
np.ones((height, width, 3)) * np.array([127.5, 127.5, 127.5]))
expand_img = Image.fromarray(expand_img)
expand_img.paste(img, (int(w_off), int(h_off)))
bbox_labels = transform_labels(bbox_labels, expand_bbox)
return expand_img, bbox_labels, width, height
return img, bbox_labels, img_width, img_height
def loadImages(self, path):
#load negative training images
def loadNegative():
negative_list = []
for dirs in self.neg_dirs:
for img_path in listdir(dirs):
base, ext = os.path.splitext(img_path)
if ext == '.jpg':
img = Image.open(dirs + '/' + img_path)
negative_list.append(img)
if self.args.verbose:
print('appending {} to {} list'.format(
img_path, 'negative'))
return negative_list
# return array of positive images
imagesList = listdir(path)
posImages, faces_bbox = [], []
left_bbox, right_bbox = [], []
#bounding boxes
dict_combo = bbox()
faces_dict, left_dict, right_dict = dict_combo[0], dict_combo[
1], dict_combo[2]
#load serially to ensure labels match
for image in imagesList:
img = Image.open(path + image)
face = faces_dict[image]
left = left_dict[image]
right = right_dict[image]
posImages.append(img)
faces_bbox.append(face)
left_bbox.append(left)
right_bbox.append(right)
negImages = loadNegative()
return posImages, negImages, faces_bbox, left_bbox, right_bbox
def testBbox(n=10):
cap = cv.VideoCapture(0)
img_seq = []
for i in range(n):
_, frame = cap.read()
img_seq.append(frame)
img_seq = np.stack(img_seq)
(frames_idx, bbox) = ut.bbox(img_seq)
n = 0
for i, img in enumerate(img_seq):
if i not in frames_idx:
print('could not find any person in frame number ' + str(i) + '!')
n += 1
continue
b = bbox[i - n]
(y0, y1, x0, x1) = [int(p) for p in b]
cut = img[y0:y1, x0:x1]
ut.imgDisplayCV(cut, "cut")
cv.waitKey(0)
return
def loadImages(self, path):
# return array of images
imagesList = listdir(path)
loadedImages, faces_bbox = [], []
left_bbox, right_bbox = [], []
#bounding boxes
dict_combo = bbox()
faces_dict, left_dict, right_dict = dict_combo[0], dict_combo[1], dict_combo[2]
#load serially to ensure labels match
for image in imagesList:
img = Image.open(path + image)
face = faces_dict[image]
left = left_dict[image]
right = right_dict[image]
loadedImages.append(img)
faces_bbox.append(face)
left_bbox.append(left)
right_bbox.append(right)
return loadedImages, faces_bbox, left_bbox, right_bbox