zdou0830 commited on
Commit
4798672
1 Parent(s): 535364e
app.py CHANGED
@@ -85,8 +85,10 @@ gr.Interface(
85
  ),
86
  ],
87
  examples=[
88
- ["./1.jpg", "a train besides sidewalk", "train;sidewalk"],
89
- ["./2.jpg", "Bumblebee, yellow with black accents.", "Bumblebee"],
 
 
90
  ],
91
  article=Path("docs/intro.md").read_text()
92
  ).launch()
 
85
  ),
86
  ],
87
  examples=[
88
+ ["./1.jpg", "A clown making a balloon animal for a pretty lady.", "clown"],
89
+ ["./1.jpg", "A clown kicking a soccer ball for a pretty lady.", "clown"],
90
+ ["./2.jpg", "A kind of tool, wooden handle with a round head.", "tool"]
91
+ ["./3.jpg", "Bumblebee, yellow with black accents.", "Bumblebee"],
92
  ],
93
  article=Path("docs/intro.md").read_text()
94
  ).launch()
maskrcnn_benchmark/engine/predictor_glip.py CHANGED
@@ -16,6 +16,7 @@ from maskrcnn_benchmark.structures.bounding_box import BoxList
16
  from maskrcnn_benchmark import layers as L
17
  from maskrcnn_benchmark.modeling.roi_heads.mask_head.inference import Masker
18
  from maskrcnn_benchmark.utils import cv2_util
 
19
 
20
  engine = inflect.engine()
21
  nltk.download("punkt")
@@ -146,6 +147,13 @@ class GLIPDemo(object):
146
  top_predictions = self._post_process(predictions, thresh)
147
 
148
  result = original_image.copy()
 
 
 
 
 
 
 
149
  if self.show_mask_heatmaps:
150
  return self.create_mask_montage(result, top_predictions)
151
 
@@ -261,6 +269,7 @@ class GLIPDemo(object):
261
  for box, color in zip(boxes, colors):
262
  box = box.to(torch.int64)
263
  top_left, bottom_right = box[:2].tolist(), box[2:].tolist()
 
264
  new_image = cv2.rectangle(new_image, tuple(top_left), tuple(bottom_right), tuple(color) if override_color is None else tuple(override_color), box_pixel)
265
 
266
  image = cv2.addWeighted(new_image, box_alpha, image, 1 - box_alpha, 0)
@@ -328,6 +337,12 @@ class GLIPDemo(object):
328
 
329
  if int(y) - text_offset_original < 20:
330
  y += 50
 
 
 
 
 
 
331
  cv2.putText(
332
  image, s, (int(x), int(y)-text_offset_original),
333
  cv2.FONT_HERSHEY_SIMPLEX, text_size,
 
16
  from maskrcnn_benchmark import layers as L
17
  from maskrcnn_benchmark.modeling.roi_heads.mask_head.inference import Masker
18
  from maskrcnn_benchmark.utils import cv2_util
19
+ from PIL import Image
20
 
21
  engine = inflect.engine()
22
  nltk.download("punkt")
 
147
  top_predictions = self._post_process(predictions, thresh)
148
 
149
  result = original_image.copy()
150
+ def resize_image_by_height(img, new_height=500):
151
+ width, height = img.size
152
+ aspect_ratio = width / height
153
+ new_width = int(new_height * aspect_ratio)
154
+ resized_img = img.resize((new_width, new_height), Image.ANTIALIAS)
155
+ return resized_img
156
+ result = resize_image_by_height(result)
157
  if self.show_mask_heatmaps:
158
  return self.create_mask_montage(result, top_predictions)
159
 
 
269
  for box, color in zip(boxes, colors):
270
  box = box.to(torch.int64)
271
  top_left, bottom_right = box[:2].tolist(), box[2:].tolist()
272
+ new_image = cv2.rectangle(new_image, tuple(top_left), tuple(bottom_right), tuple(255, 255, 255), box_pixel+3)
273
  new_image = cv2.rectangle(new_image, tuple(top_left), tuple(bottom_right), tuple(color) if override_color is None else tuple(override_color), box_pixel)
274
 
275
  image = cv2.addWeighted(new_image, box_alpha, image, 1 - box_alpha, 0)
 
337
 
338
  if int(y) - text_offset_original < 20:
339
  y += 50
340
+ text_size = cv2.getTextSize(s, cv2.FONT_HERSHEY_SIMPLEX, text_size, text_pixel)[0]
341
+ position = (int(x), int(y)-text_offset_original)
342
+ bottom_left_corner_of_text = (position[0], position[1] + text_size[1])
343
+ top_right_corner = (position[0] + text_size[0], position[1])
344
+
345
+ image[position[1]:bottom_left_corner_of_text[1], position[0]:top_right_corner[0]] = (255, 255, 255)
346
  cv2.putText(
347
  image, s, (int(x), int(y)-text_offset_original),
348
  cv2.FONT_HERSHEY_SIMPLEX, text_size,