ivelin commited on
Commit
c04ab4e
1 Parent(s): a8e6b52

fix: gradio update related issues

Browse files

Signed-off-by: ivelin <ivelin.eth@gmail.com>

Files changed (1) hide show
  1. app.py +7 -6
app.py CHANGED
@@ -15,7 +15,8 @@ def load_model(pretrained_revision: str = 'main'):
15
  pretrained_repo_name = 'ivelin/donut-refexp-click'
16
  # revision can be git commit hash, branch or tag
17
  # use 'main' for latest revision
18
- print(f"Loading model checkpoint from repo: {pretrained_repo_name}, revision: {pretrained_revision}")
 
19
  processor = DonutProcessor.from_pretrained(
20
  pretrained_repo_name, revision=pretrained_revision, use_auth_token="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")
21
  processor.image_processor.do_align_long_axis = False
@@ -28,10 +29,10 @@ def load_model(pretrained_revision: str = 'main'):
28
  print(f'processor image size: {processor.image_processor.size}')
29
  model = VisionEncoderDecoderModel.from_pretrained(
30
  pretrained_repo_name, use_auth_token="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", revision=pretrained_revision)
31
-
32
  device = "cuda" if torch.cuda.is_available() else "cpu"
33
  model.to(device)
34
-
35
 
36
  def prepare_image_for_encoder(image=None, output_image_size=None):
37
  """
@@ -89,13 +90,13 @@ def translate_point_coords_from_out_to_in(point=None, input_image_size=None, out
89
  f"translated point={point}, resized_image_size: {resized_width, resized_height}")
90
 
91
 
92
- def process_refexp(image: Image, prompt: str, model_revision: str = 'main'):
93
 
94
  print(f"(image, prompt): {image}, {prompt}")
95
 
96
  if not model_revision:
97
  model_revision = 'main'
98
-
99
  print(f"model checkpoint revision: {model_revision}")
100
 
101
  load_model(model_revision)
@@ -222,7 +223,7 @@ examples = [["example_1.jpg", "select the setting icon from top right corner", "
222
 
223
  demo = gr.Interface(fn=process_refexp,
224
  inputs=[gr.Image(type="pil"), "text", "text"],
225
- outputs=[gr.Image(type="pil"), "json"],
226
  title=title,
227
  description=description,
228
  article=article,
 
15
  pretrained_repo_name = 'ivelin/donut-refexp-click'
16
  # revision can be git commit hash, branch or tag
17
  # use 'main' for latest revision
18
+ print(
19
+ f"Loading model checkpoint from repo: {pretrained_repo_name}, revision: {pretrained_revision}")
20
  processor = DonutProcessor.from_pretrained(
21
  pretrained_repo_name, revision=pretrained_revision, use_auth_token="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")
22
  processor.image_processor.do_align_long_axis = False
 
29
  print(f'processor image size: {processor.image_processor.size}')
30
  model = VisionEncoderDecoderModel.from_pretrained(
31
  pretrained_repo_name, use_auth_token="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", revision=pretrained_revision)
32
+
33
  device = "cuda" if torch.cuda.is_available() else "cpu"
34
  model.to(device)
35
+
36
 
37
  def prepare_image_for_encoder(image=None, output_image_size=None):
38
  """
 
90
  f"translated point={point}, resized_image_size: {resized_width, resized_height}")
91
 
92
 
93
+ def process_refexp(image, prompt: str, model_revision: str = 'main'):
94
 
95
  print(f"(image, prompt): {image}, {prompt}")
96
 
97
  if not model_revision:
98
  model_revision = 'main'
99
+
100
  print(f"model checkpoint revision: {model_revision}")
101
 
102
  load_model(model_revision)
 
223
 
224
  demo = gr.Interface(fn=process_refexp,
225
  inputs=[gr.Image(type="pil"), "text", "text"],
226
+ outputs=[gr.Image(type="pil", shape=(820)), "json"],
227
  title=title,
228
  description=description,
229
  article=article,