def fetch_image(self):
# load
if os.path.exists(self.image_path) is not True:
raise IOError("File does not exist: %s" % self.image_path)
img = load_image(self.image_path)
self.origin_height = img.shape[0]
self.origin_width = img.shape[1]
# resize
if img.shape[0] > img.shape[1]:
scale = 1.0 * self.max_edge / img.shape[0]
else:
scale = 1.0 * self.max_edge / img.shape[1]
h, w = int(self.origin_height * scale), int(self.origin_width * scale)
h, w = h - h%16, w - w%16
img = cv2.resize(img, (w, h), interpolation=cv2.INTER_LINEAR)
img = sub_mean(img)
if self.flipped:
img = img[:, ::-1, :]
self.height = img.shape[0]
self.width = img.shape[1]
self.image = img
self.img_blob = self.image.transpose((2, 0, 1))
self.img_blob = self.img_blob[np.newaxis, ...]
self.scale = scale
self.objAttMaskScales = []
return {'image': self.img_blob}
def fetch_image(self):
# load
if os.path.exists(self.image_path) is not True:
raise IOError("File does not exist: %s" % self.image_path)
img = load_image(self.image_path)
self.origin_height = img.shape[0]
self.origin_width = img.shape[1]
# resize
if img.shape[0] > img.shape[1]:
scale = 1.0 * self.max_edge / img.shape[0]
else:
scale = 1.0 * self.max_edge / img.shape[1]
h, w = int(self.origin_height * scale), int(self.origin_width * scale)
# make sure that w and h
# could be divisible by 4
# so that we can get a
# predictable sized feature
# map from body net.
h, w = h - h % 4, w - w % 4
img = cv2.resize(img, (w, h), interpolation=cv2.INTER_LINEAR)
img = sub_mean(img)
if self.flipped:
img = img[:, ::-1, :]
self.height = img.shape[0]
self.width = img.shape[1]
self.image = img
self.img_blob = self.image.transpose((2, 0, 1))
self.img_blob = self.img_blob[np.newaxis, ...]
self.scale = scale
return {'image': self.img_blob}
# image pre-processing
img = load_image(args.input_image)
img_org = img.copy()[:,:,::-1]
origin_height = img.shape[0]
origin_width = img.shape[1]
if img.shape[0] > img.shape[1]:
scale = 1.0 * config.TEST_SCALE / img.shape[0]
else:
scale = 1.0 * config.TEST_SCALE / img.shape[1]
h, w = int(origin_height * scale), int(origin_width * scale)
h, w = h - h % 16, w - w % 16
img = cv2.resize(img, (w, h), interpolation=cv2.INTER_LINEAR)
img = sub_mean(img)
img_blob = img.transpose((2, 0, 1))
img_blob = img_blob[np.newaxis, ...]
# all masks with all scores (1000 in total)
ret_masks, ret_scores = gen_masks_new(net, img_blob, config, dest_shape=(origin_height, origin_width))
# display top 20 masks
for i in np.argsort(ret_scores)[::-1][:20]:
mask = ret_masks[i].copy()
color = np.array(colors[i%len(colors)])
mask = np.dstack([mask]*3)
colorMask = color*mask
img_org=img_org*(1-mask)+mask*img_org*.5+colorMask*.5
img_org = mark_boundaries(img_org,mask[:,:,0],color=color.tolist(),mode='thick')