def check_sanity(self):
for i in tqdm(self.image_ids):
assert self.image_files[i][:-4] == self.label_files[i][:-4],\
'image - label filename mismatch: {} - {}'.format(self.image_files[i], self.label_files[i])
img = load_image_rgb(
os.path.join(self.image_dir, self.image_files[i]))
msk = load_image_rgb(
os.path.join(self.label_dir, self.label_files[i]))
assert img.shape == msk.shape,\
'img.shape: {}, msk.shape: {}'.format(img.shape, msk.shape)
def display(self):
should_update = super(GuiInferenceViewer, self).display()
if should_update:
image_file = self.image_files[self.image_id]
print('Processing {}...'.format(os.path.basename(image_file)))
img = load_image_rgb(image_file)
t = time.time()
det = self.model.predict(img.astype(np.float32), self.threshold)
time_taken = time.time() - t
print("\ttime taken: {}".format(time_taken))
if self.model.config.MODEL == 'maskrcnn':
display_instances(image=img,
boxes=det['rois'],
masks=det['masks'],
class_ids=det['class_ids'],
scores=det['scores'],
title=image_file,
class_names=self.class_names,
ax=self.ax)
else:
display_segmentation(image=img,
masks=det,
class_names=self.class_names,
ax=self.ax)
title = "ID: {}\nImage file name: {}".format(
self.image_id, os.path.basename(image_file))
self.fig.suptitle(title, fontsize=20)
self.fig.canvas.draw_idle()
def display(self):
should_update = super(GuiImageViewer, self).display()
if should_update:
image_file = self.image_files[self.image_id]
print('Processing {}...'.format(os.path.basename(image_file)))
img = load_image_rgb(image_file)
self.ax.imshow(img.astype(np.uint8))
title = "ID: {}\nImage file name: {}".format(
self.image_id, os.path.basename(image_file))
self.fig.suptitle(title, fontsize=20)
self.fig.canvas.draw_idle()
def load_mask(self, image_id):
rgb_mask = load_image_rgb(
os.path.join(self.label_dir, self.label_files[image_id]))
mask = np.zeros(
(rgb_mask.shape[0], rgb_mask.shape[1], self.num_classes - 1))
for cls in range(self.num_classes - 1):
colors = self.trainId2colors[cls]
cls_mask = np.zeros((rgb_mask.shape[0], rgb_mask.shape[1]))
for color in colors:
cls_mask = np.logical_or(cls_mask,
(rgb_mask == color).all(axis=2))
mask[:, :, cls] = cls_mask
return mask
model.load_weights(args.weights)
image_files = sorted([
os.path.join(args.image_dir, x) for x in os.listdir(args.image_dir)
if x.lower().endswith('.jpg') or x.lower().endswith('.png')
or x.lower().endswith('.bmp')
])
dst_dir = os.path.join(args.image_dir, 'results')
if not os.path.exists(dst_dir):
os.mkdir(dst_dir)
for image_file in image_files:
filename = os.path.basename(image_file)
print('Processing {}...'.format(filename))
img = load_image_rgb(image_file)
t = time.time()
det = model.predict(img.astype(np.float32), args.threshold)
time_taken = time.time() - t
print("\ttime taken: {}".format(time_taken))
if model.config.MODEL == 'maskrcnn':
res = draw_instances(image=img,
boxes=det['rois'],
masks=det['masks'],
class_ids=det['class_ids'],
scores=det['scores'],
title=image_file,
class_names=label)
else:
res = draw_segmentation(image=img, masks=det, class_names=label)
def load_image(self, image_id):
return load_image_rgb(
os.path.join(self.image_dir, self.image_files[image_id]))
def load_image(self, image_id):
return load_image_rgb(self.annotations[image_id]['path'])
def load_image(self, image_id):
image = load_image_rgb(self.image_info[image_id]['path'])
return image
