Spaces:
Running
Running
NBoukachab
commited on
Commit
•
b858330
1
Parent(s):
a568da9
Add a model selector
Browse files- app.py +78 -48
- config.json +0 -10
- config.py +10 -9
- config.yaml +30 -0
- resource/hugging_face_1.jpg +0 -0
- resource/hugging_face_2.jpg +0 -0
- resource/hugging_face_3.jpg +0 -0
- resource/hugging_face_4.jpg +0 -0
- tools.py +49 -0
app.py
CHANGED
@@ -1,51 +1,59 @@
|
|
1 |
# -*- coding: utf-8 -*-
|
2 |
|
3 |
import json
|
4 |
-
import os
|
5 |
from pathlib import Path
|
6 |
|
7 |
import gradio as gr
|
8 |
import numpy as np
|
9 |
-
from doc_ufcn import models
|
10 |
-
from doc_ufcn.main import DocUFCN
|
11 |
from PIL import Image, ImageDraw
|
12 |
|
13 |
from config import parse_configurations
|
|
|
14 |
|
15 |
# Load the config
|
16 |
-
config = parse_configurations(Path("config.
|
17 |
|
18 |
-
#
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
23 |
|
24 |
-
#
|
25 |
-
|
26 |
|
27 |
-
# Check that the number of colors is equal to the number of classes -1
|
28 |
-
assert len(classes) - 1 == len(
|
29 |
-
classes_colors
|
30 |
-
), f"The parameter classes_colors was filled with the wrong number of colors. {len(classes)-1} colors are expected instead of {len(classes_colors)}."
|
31 |
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
model.
|
|
|
|
|
43 |
|
44 |
|
45 |
-
def query_image(image):
|
46 |
"""
|
47 |
-
|
48 |
|
|
|
49 |
:param image: An image to predict
|
50 |
:return: Image and dict, an image with the predictions and a
|
51 |
dictionary mapping an object idx (starting from 1) to a dictionary describing the detected object:
|
@@ -54,8 +62,11 @@ def query_image(image):
|
|
54 |
- `channel` key : str, the name of the predicted class.
|
55 |
"""
|
56 |
|
|
|
|
|
|
|
57 |
# Make a prediction with the model
|
58 |
-
detected_polygons, probabilities, mask, overlap = model.predict(
|
59 |
input_image=image, raw_output=True, mask_output=True, overlap_output=True
|
60 |
)
|
61 |
|
@@ -70,12 +81,12 @@ def query_image(image):
|
|
70 |
|
71 |
# Create the polygons on the copy of the image for each class with the corresponding color
|
72 |
# We do not draw polygons of the background channel (channel 0)
|
73 |
-
for channel in range(1,
|
74 |
for i, polygon in enumerate(detected_polygons[channel]):
|
75 |
# Draw the polygons on the image copy.
|
76 |
# Loop through the class_colors list (channel 1 has color 0)
|
77 |
ImageDraw.Draw(img2).polygon(
|
78 |
-
polygon["polygon"], fill=
|
79 |
)
|
80 |
|
81 |
# Build the dictionary
|
@@ -88,34 +99,55 @@ def query_image(image):
|
|
88 |
# Confidence that the model predicts the polygon in the right place
|
89 |
"confidence": polygon["confidence"],
|
90 |
# The channel on which the polygon is predicted
|
91 |
-
"channel": classes[channel],
|
92 |
}
|
93 |
)
|
94 |
|
95 |
# Return the blend of the images and the dictionary formatted in json
|
96 |
-
return Image.blend(image, img2, 0.5), json.dumps(predict, indent=
|
97 |
|
98 |
|
99 |
-
|
|
|
|
|
100 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
101 |
# Create app title
|
102 |
gr.Markdown(f"# {config['title']}")
|
103 |
|
104 |
# Create app description
|
105 |
gr.Markdown(config["description"])
|
106 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
107 |
# Create a first row of blocks
|
108 |
with gr.Row():
|
109 |
-
|
110 |
# Create a column on the left
|
111 |
with gr.Column():
|
112 |
-
|
113 |
# Generates an image that can be uploaded by a user
|
114 |
image = gr.Image()
|
115 |
|
116 |
# Create a row under the image
|
117 |
with gr.Row():
|
118 |
-
|
119 |
# Generate a button to clear the inputs and outputs
|
120 |
clear_button = gr.Button("Clear", variant="secondary")
|
121 |
|
@@ -124,25 +156,21 @@ with gr.Blocks() as process_image:
|
|
124 |
|
125 |
# Create a row under the buttons
|
126 |
with gr.Row():
|
127 |
-
|
128 |
-
|
129 |
-
examples = gr.Examples(inputs=image, examples=config["examples"])
|
130 |
|
131 |
# Create a column on the right
|
132 |
with gr.Column():
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
|
137 |
# Create a row under the predicted image
|
138 |
with gr.Row():
|
139 |
-
|
140 |
# Create a column so that the JSON output doesn't take the full size of the page
|
141 |
with gr.Column():
|
142 |
-
|
143 |
-
# Create a collapsible region
|
144 |
with gr.Accordion("JSON"):
|
145 |
-
|
146 |
# Generates a json with the model predictions
|
147 |
json_output = gr.JSON()
|
148 |
|
@@ -154,7 +182,9 @@ with gr.Blocks() as process_image:
|
|
154 |
)
|
155 |
|
156 |
# Create the button to submit the prediction
|
157 |
-
submit_button.click(
|
|
|
|
|
158 |
|
159 |
-
# Launch the application
|
160 |
process_image.launch()
|
|
|
1 |
# -*- coding: utf-8 -*-
|
2 |
|
3 |
import json
|
|
|
4 |
from pathlib import Path
|
5 |
|
6 |
import gradio as gr
|
7 |
import numpy as np
|
|
|
|
|
8 |
from PIL import Image, ImageDraw
|
9 |
|
10 |
from config import parse_configurations
|
11 |
+
from tools import UFCNModel
|
12 |
|
13 |
# Load the config
|
14 |
+
config = parse_configurations(Path("config.yaml"))
|
15 |
|
16 |
+
# Check that the paths of the examples are valid
|
17 |
+
for example in config["examples"]:
|
18 |
+
assert Path.exists(
|
19 |
+
Path(example)
|
20 |
+
), f"The path of the image '{example}' does not exist."
|
21 |
+
|
22 |
+
# Cached models, maps model_name to UFCNModel object
|
23 |
+
MODELS = {
|
24 |
+
model["model_name"]: UFCNModel(
|
25 |
+
name=model["model_name"],
|
26 |
+
colors=model["classes_colors"],
|
27 |
+
title=model["title"],
|
28 |
+
description=model["description"],
|
29 |
+
)
|
30 |
+
for model in config["models"]
|
31 |
+
}
|
32 |
|
33 |
+
# Create a list of models name
|
34 |
+
models_name = list(MODELS)
|
35 |
|
|
|
|
|
|
|
|
|
36 |
|
37 |
+
def load_model(model_name) -> UFCNModel:
|
38 |
+
"""
|
39 |
+
Retrieve the model, and load its parameters/files if it wasn't done before.
|
40 |
|
41 |
+
:param model_name: The name of the selected model
|
42 |
+
:return: The UFCNModel instance selected
|
43 |
+
"""
|
44 |
+
assert model_name in MODELS
|
45 |
+
model = MODELS[model_name]
|
46 |
+
# Load the model's files if it wasn't done before
|
47 |
+
if not model.loaded:
|
48 |
+
model.load()
|
49 |
+
return model
|
50 |
|
51 |
|
52 |
+
def query_image(model_name: gr.Dropdown, image: gr.Image) -> list([Image, json]):
|
53 |
"""
|
54 |
+
Loads a model and draws the predicted polygons with the color provided by the model on an image
|
55 |
|
56 |
+
:param model: A model selected in dropdown
|
57 |
:param image: An image to predict
|
58 |
:return: Image and dict, an image with the predictions and a
|
59 |
dictionary mapping an object idx (starting from 1) to a dictionary describing the detected object:
|
|
|
62 |
- `channel` key : str, the name of the predicted class.
|
63 |
"""
|
64 |
|
65 |
+
# Load the model and get its classes, classes_colors and the model
|
66 |
+
ufcn_model = load_model(model_name)
|
67 |
+
|
68 |
# Make a prediction with the model
|
69 |
+
detected_polygons, probabilities, mask, overlap = ufcn_model.model.predict(
|
70 |
input_image=image, raw_output=True, mask_output=True, overlap_output=True
|
71 |
)
|
72 |
|
|
|
81 |
|
82 |
# Create the polygons on the copy of the image for each class with the corresponding color
|
83 |
# We do not draw polygons of the background channel (channel 0)
|
84 |
+
for channel in range(1, ufcn_model.num_channels):
|
85 |
for i, polygon in enumerate(detected_polygons[channel]):
|
86 |
# Draw the polygons on the image copy.
|
87 |
# Loop through the class_colors list (channel 1 has color 0)
|
88 |
ImageDraw.Draw(img2).polygon(
|
89 |
+
polygon["polygon"], fill=ufcn_model.colors[channel - 1]
|
90 |
)
|
91 |
|
92 |
# Build the dictionary
|
|
|
99 |
# Confidence that the model predicts the polygon in the right place
|
100 |
"confidence": polygon["confidence"],
|
101 |
# The channel on which the polygon is predicted
|
102 |
+
"channel": ufcn_model.classes[channel],
|
103 |
}
|
104 |
)
|
105 |
|
106 |
# Return the blend of the images and the dictionary formatted in json
|
107 |
+
return Image.blend(image, img2, 0.5), json.dumps(predict, indent=2)
|
108 |
|
109 |
|
110 |
+
def update_model(model_name: gr.Dropdown) -> str:
|
111 |
+
"""
|
112 |
+
Update the model title to the title of the current model
|
113 |
|
114 |
+
:param model_name: The name of the selected model
|
115 |
+
:return: A new title
|
116 |
+
"""
|
117 |
+
return f"## {MODELS[model_name].title}", MODELS[model_name].description
|
118 |
+
|
119 |
+
|
120 |
+
with gr.Blocks() as process_image:
|
121 |
# Create app title
|
122 |
gr.Markdown(f"# {config['title']}")
|
123 |
|
124 |
# Create app description
|
125 |
gr.Markdown(config["description"])
|
126 |
|
127 |
+
# Create dropdown button
|
128 |
+
model_name = gr.Dropdown(models_name, value=models_name[0], label="Models")
|
129 |
+
|
130 |
+
# get models
|
131 |
+
selected_model: UFCNModel = MODELS[model_name.value]
|
132 |
+
|
133 |
+
# Create model title
|
134 |
+
model_title = gr.Markdown(f"## {selected_model.title}")
|
135 |
+
|
136 |
+
# Create model description
|
137 |
+
model_description = gr.Markdown(selected_model.description)
|
138 |
+
|
139 |
+
# Change model title and description when the model_id is update
|
140 |
+
model_name.change(update_model, model_name, [model_title, model_description])
|
141 |
+
|
142 |
# Create a first row of blocks
|
143 |
with gr.Row():
|
|
|
144 |
# Create a column on the left
|
145 |
with gr.Column():
|
|
|
146 |
# Generates an image that can be uploaded by a user
|
147 |
image = gr.Image()
|
148 |
|
149 |
# Create a row under the image
|
150 |
with gr.Row():
|
|
|
151 |
# Generate a button to clear the inputs and outputs
|
152 |
clear_button = gr.Button("Clear", variant="secondary")
|
153 |
|
|
|
156 |
|
157 |
# Create a row under the buttons
|
158 |
with gr.Row():
|
159 |
+
# Generate example images that can be used as input image for every model
|
160 |
+
gr.Examples(config["examples"], inputs=image)
|
|
|
161 |
|
162 |
# Create a column on the right
|
163 |
with gr.Column():
|
164 |
+
with gr.Row():
|
165 |
+
# Generates an output image that does not support upload
|
166 |
+
image_output = gr.Image(interactive=False)
|
167 |
|
168 |
# Create a row under the predicted image
|
169 |
with gr.Row():
|
|
|
170 |
# Create a column so that the JSON output doesn't take the full size of the page
|
171 |
with gr.Column():
|
172 |
+
# # Create a collapsible region
|
|
|
173 |
with gr.Accordion("JSON"):
|
|
|
174 |
# Generates a json with the model predictions
|
175 |
json_output = gr.JSON()
|
176 |
|
|
|
182 |
)
|
183 |
|
184 |
# Create the button to submit the prediction
|
185 |
+
submit_button.click(
|
186 |
+
query_image, inputs=[model_name, image], outputs=[image_output, json_output]
|
187 |
+
)
|
188 |
|
189 |
+
# Launch the application with the public mode (True or False)
|
190 |
process_image.launch()
|
config.json
DELETED
@@ -1,10 +0,0 @@
|
|
1 |
-
{
|
2 |
-
"model_name": "doc-ufcn-generic-historical-line",
|
3 |
-
"classes_colors": ["green"],
|
4 |
-
"title":"doc-ufcn Line Detection Demo",
|
5 |
-
"description":"A demo showing a prediction from the [Teklia/doc-ufcn-generic-historical-line](https://huggingface.co/Teklia/doc-ufcn-generic-historical-line) model. The generic historical line detection model predicts text lines from document images.",
|
6 |
-
"examples":[
|
7 |
-
"resource/hugging_face_1.jpg",
|
8 |
-
"resource/hugging_face_2.jpg"
|
9 |
-
]
|
10 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
config.py
CHANGED
@@ -7,21 +7,22 @@ from teklia_toolbox.config import ConfigParser
|
|
7 |
|
8 |
def parse_configurations(config_path: Path):
|
9 |
"""
|
10 |
-
Parse multiple
|
11 |
of configuration for the HuggingFace app
|
12 |
|
13 |
-
:param config_path: pathlib.Path, Path to the .
|
14 |
:return: dict, containing the configuration. Ensures config is complete and with correct typing
|
15 |
"""
|
16 |
-
|
17 |
parser = ConfigParser()
|
18 |
|
19 |
-
parser.add_option(
|
20 |
-
|
21 |
-
)
|
22 |
-
parser.add_option("classes_colors", type=list, default=["green"])
|
23 |
-
parser.add_option("title", type=str)
|
24 |
-
parser.add_option("description", type=str)
|
25 |
parser.add_option("examples", type=list)
|
|
|
|
|
|
|
|
|
|
|
|
|
26 |
|
27 |
return parser.parse(config_path)
|
|
|
7 |
|
8 |
def parse_configurations(config_path: Path):
|
9 |
"""
|
10 |
+
Parse multiple YAML configuration files into a single source
|
11 |
of configuration for the HuggingFace app
|
12 |
|
13 |
+
:param config_path: pathlib.Path, Path to the .yaml config file
|
14 |
:return: dict, containing the configuration. Ensures config is complete and with correct typing
|
15 |
"""
|
|
|
16 |
parser = ConfigParser()
|
17 |
|
18 |
+
parser.add_option("title")
|
19 |
+
parser.add_option("description")
|
|
|
|
|
|
|
|
|
20 |
parser.add_option("examples", type=list)
|
21 |
+
model_parser = parser.add_subparser("models", many=True)
|
22 |
+
|
23 |
+
model_parser.add_option("model_name")
|
24 |
+
model_parser.add_option("title")
|
25 |
+
model_parser.add_option("description")
|
26 |
+
model_parser.add_option("classes_colors", type=list)
|
27 |
|
28 |
return parser.parse(config_path)
|
config.yaml
ADDED
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
title: Teklia - Doc-UFCN Demo
|
3 |
+
description: >-
|
4 |
+
[TEKLIA](https://teklia.com/)’s Document Layout Analysis on historical documents. For modern documents, see [ocelus.teklia.com](https://ocelus.teklia.com).
|
5 |
+
examples:
|
6 |
+
- resource/hugging_face_1.jpg
|
7 |
+
- resource/hugging_face_2.jpg
|
8 |
+
- resource/hugging_face_3.jpg
|
9 |
+
- resource/hugging_face_4.jpg
|
10 |
+
|
11 |
+
models:
|
12 |
+
- model_name: doc-ufcn-generic-historical-line
|
13 |
+
title: Doc-UFCN Generic historical line detection
|
14 |
+
description: >-
|
15 |
+
The [generic historical line detection model](https://huggingface.co/Teklia/doc-ufcn-generic-historical-line) predicts text lines from document images. Please select an image from the examples below or upload your own image!
|
16 |
+
classes_colors:
|
17 |
+
- green
|
18 |
+
- model_name: doc-ufcn-huginmunin-line
|
19 |
+
title: Doc-UFCN Hugin-Munin line detection
|
20 |
+
description: >-
|
21 |
+
The [Hugin-Munin line detection model](https://huggingface.co/Teklia/doc-ufcn-huginmunin-line) predicts horizontal and vertical text lines from Hugin-Munin document images. Please select an image from the examples below or upload your own image!
|
22 |
+
classes_colors:
|
23 |
+
- green
|
24 |
+
- blue
|
25 |
+
- model_name: doc-ufcn-generic-page
|
26 |
+
title: Doc-UFCN Generic page detection
|
27 |
+
description: >-
|
28 |
+
The [generic page detection model](https://huggingface.co/Teklia/doc-ufcn-generic-page) predicts single pages from document images. Please select an image from the examples below or upload your own image!
|
29 |
+
classes_colors:
|
30 |
+
- green
|
resource/hugging_face_1.jpg
CHANGED
resource/hugging_face_2.jpg
CHANGED
resource/hugging_face_3.jpg
ADDED
resource/hugging_face_4.jpg
ADDED
tools.py
ADDED
@@ -0,0 +1,49 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# -*- coding: utf-8 -*-
|
2 |
+
|
3 |
+
from dataclasses import dataclass, field
|
4 |
+
|
5 |
+
from doc_ufcn import models
|
6 |
+
from doc_ufcn.main import DocUFCN
|
7 |
+
|
8 |
+
|
9 |
+
@dataclass
|
10 |
+
class UFCNModel:
|
11 |
+
name: str
|
12 |
+
colors: list
|
13 |
+
title: str
|
14 |
+
description: str
|
15 |
+
classes: list = field(default_factory=list)
|
16 |
+
model: DocUFCN = None
|
17 |
+
|
18 |
+
def get_class_name(self, channel_idx):
|
19 |
+
return self.classes[channel_idx]
|
20 |
+
|
21 |
+
@property
|
22 |
+
def loaded(self):
|
23 |
+
return self.model is not None
|
24 |
+
|
25 |
+
@property
|
26 |
+
def num_channels(self):
|
27 |
+
return len(self.classes)
|
28 |
+
|
29 |
+
def load(self):
|
30 |
+
# Download the model
|
31 |
+
model_path, parameters = models.download_model(name=self.name)
|
32 |
+
|
33 |
+
# Store classes
|
34 |
+
self.classes = parameters["classes"]
|
35 |
+
|
36 |
+
# Check that the number of colors is equal to the number of classes -1
|
37 |
+
assert self.num_channels - 1 == len(
|
38 |
+
self.colors
|
39 |
+
), f"The parameter classes_colors was filled with the wrong number of colors. {self.num_channels-1} colors are expected instead of {len(self.colors)}."
|
40 |
+
|
41 |
+
# Load the model
|
42 |
+
self.model = DocUFCN(
|
43 |
+
no_of_classes=len(self.classes),
|
44 |
+
model_input_size=parameters["input_size"],
|
45 |
+
device="cpu",
|
46 |
+
)
|
47 |
+
self.model.load(
|
48 |
+
model_path=model_path, mean=parameters["mean"], std=parameters["std"]
|
49 |
+
)
|