Charlie Li
commited on
Commit
β’
4697797
1
Parent(s):
f44710a
build all
Browse files- .gitignore +7 -0
- README.md +2 -2
- app.py +101 -0
- org/cor.svg +264 -0
- requirements.txt +5 -0
- utils.py +235 -0
.gitignore
ADDED
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
__pycache__
|
2 |
+
*.mp4
|
3 |
+
flagged/
|
4 |
+
derendering_supp/
|
5 |
+
*.zip
|
6 |
+
__MACOSX/
|
7 |
+
.DS_Store
|
README.md
CHANGED
@@ -1,6 +1,6 @@
|
|
1 |
---
|
2 |
title: Model Output Playground
|
3 |
-
emoji:
|
4 |
colorFrom: purple
|
5 |
colorTo: green
|
6 |
sdk: gradio
|
@@ -10,4 +10,4 @@ pinned: false
|
|
10 |
license: apache-2.0
|
11 |
---
|
12 |
|
13 |
-
|
|
|
1 |
---
|
2 |
title: Model Output Playground
|
3 |
+
emoji: π
|
4 |
colorFrom: purple
|
5 |
colorTo: green
|
6 |
sdk: gradio
|
|
|
10 |
license: apache-2.0
|
11 |
---
|
12 |
|
13 |
+
Paper: https://arxiv.org/abs/2402.05804
|
app.py
ADDED
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from utils import *
|
3 |
+
|
4 |
+
file_url = "https://storage.googleapis.com/derendering_model/derendering_supp.zip"
|
5 |
+
filename = "derendering_supp.zip"
|
6 |
+
|
7 |
+
download_file(file_url, filename)
|
8 |
+
unzip_file(filename)
|
9 |
+
print("Downloaded and unzipped the file.")
|
10 |
+
|
11 |
+
diagram = get_svg_content("derendering_supp/derender_diagram.svg")
|
12 |
+
org = get_svg_content("org/cor.svg")
|
13 |
+
|
14 |
+
org_content = f"""
|
15 |
+
{org}
|
16 |
+
"""
|
17 |
+
|
18 |
+
|
19 |
+
def demo(Dataset, Model):
|
20 |
+
if Model == "Small-i":
|
21 |
+
inkml_path = f"./derendering_supp/small-i_{Dataset}_inkml"
|
22 |
+
elif Model == "Small-p":
|
23 |
+
inkml_path = f"./derendering_supp/small-p_{Dataset}_inkml"
|
24 |
+
elif Model == "Large-i":
|
25 |
+
inkml_path = f"./derendering_supp/large-i_{Dataset}_inkml"
|
26 |
+
|
27 |
+
path = f"./derendering_supp/{Dataset}/images_sample"
|
28 |
+
samples = os.listdir(path)
|
29 |
+
# Randomly pick a sample
|
30 |
+
picked_samples = random.sample(samples, min(1, len(samples)))
|
31 |
+
|
32 |
+
query_modes = ["d+t", "r+d", "vanilla"]
|
33 |
+
plot_title = {"r+d": "Recognized: ", "d+t": "OCR Input: ", "vanilla": ""}
|
34 |
+
text_outputs = []
|
35 |
+
|
36 |
+
for name in picked_samples:
|
37 |
+
img_path = os.path.join(path, name)
|
38 |
+
img = load_and_pad_img_dir(img_path)
|
39 |
+
|
40 |
+
for mode in query_modes:
|
41 |
+
example_id = name.strip(".png")
|
42 |
+
inkml_file = os.path.join(inkml_path, mode, example_id + ".inkml")
|
43 |
+
text_field = parse_inkml_annotations(inkml_file)["textField"]
|
44 |
+
output_text = f"{plot_title[mode]}{text_field}"
|
45 |
+
text_outputs.append(output_text) # Append text output for the current mode
|
46 |
+
ink = inkml_to_ink(inkml_file)
|
47 |
+
plot_ink_to_video(ink, mode + ".mp4", input_image=img)
|
48 |
+
|
49 |
+
return (
|
50 |
+
img,
|
51 |
+
text_outputs[0],
|
52 |
+
"d+t.mp4",
|
53 |
+
text_outputs[1],
|
54 |
+
"r+d.mp4",
|
55 |
+
text_outputs[2],
|
56 |
+
"vanilla.mp4",
|
57 |
+
)
|
58 |
+
|
59 |
+
|
60 |
+
with gr.Blocks() as app:
|
61 |
+
gr.HTML(org_content)
|
62 |
+
gr.Markdown(
|
63 |
+
f"""
|
64 |
+
# InkSight: Offline-to-Online Handwriting Conversion by Learning to Read and Write<br>
|
65 |
+
<div>{diagram}</div>
|
66 |
+
π This demo showcases the outputs of <b>Small-i</b>, <b>Small-p</b>, and <b>Large-i</b> on three public datasets (100 samples each).<br>
|
67 |
+
βΉοΈ Choose a model variant and dataset, then click 'Sample' to see an input with its corresponding outputs for all three inference types..<br>
|
68 |
+
"""
|
69 |
+
)
|
70 |
+
with gr.Row():
|
71 |
+
dataset = gr.Dropdown(
|
72 |
+
["IMGUR5K", "IAM", "HierText"], label="Dataset", value="HierText"
|
73 |
+
)
|
74 |
+
model = gr.Dropdown(
|
75 |
+
["Small-i", "Large-i", "Small-p"],
|
76 |
+
label="InkSight Model Variant",
|
77 |
+
value="Small-i",
|
78 |
+
)
|
79 |
+
im = gr.Image(label="Input Image")
|
80 |
+
with gr.Row():
|
81 |
+
d_t_text = gr.Textbox(
|
82 |
+
label="OCR recognition input to the model", interactive=False
|
83 |
+
)
|
84 |
+
r_d_text = gr.Textbox(label="Recognition from the model", interactive=False)
|
85 |
+
vanilla_text = gr.Textbox(label="Vanilla", interactive=False)
|
86 |
+
|
87 |
+
with gr.Row():
|
88 |
+
d_t = gr.Video(label="Derender with Text", autoplay=True)
|
89 |
+
r_d = gr.Video(label="Recognize and Derender", autoplay=True)
|
90 |
+
vanilla = gr.Video(label="Vanilla", autoplay=True)
|
91 |
+
|
92 |
+
with gr.Row():
|
93 |
+
btn_sub = gr.Button("Sample")
|
94 |
+
|
95 |
+
btn_sub.click(
|
96 |
+
fn=demo,
|
97 |
+
inputs=[dataset, model],
|
98 |
+
outputs=[im, d_t_text, d_t, r_d_text, r_d, vanilla_text, vanilla],
|
99 |
+
)
|
100 |
+
|
101 |
+
app.launch()
|
org/cor.svg
ADDED
requirements.txt
ADDED
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
tqdm
|
2 |
+
numpy
|
3 |
+
matplotlib
|
4 |
+
Pillow
|
5 |
+
numpy
|
utils.py
ADDED
@@ -0,0 +1,235 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import json
|
2 |
+
from tqdm import tqdm
|
3 |
+
import numpy as np
|
4 |
+
import matplotlib.pyplot as plt
|
5 |
+
import xml.etree.ElementTree as ET
|
6 |
+
from xml.dom import minidom
|
7 |
+
import os
|
8 |
+
from PIL import Image
|
9 |
+
import matplotlib.animation as animation
|
10 |
+
import copy
|
11 |
+
from PIL import ImageEnhance
|
12 |
+
import colorsys
|
13 |
+
import matplotlib.colors as mcolors
|
14 |
+
from matplotlib.collections import LineCollection
|
15 |
+
from matplotlib.patheffects import withStroke
|
16 |
+
import random
|
17 |
+
import warnings
|
18 |
+
from matplotlib.figure import Figure
|
19 |
+
from io import BytesIO
|
20 |
+
from matplotlib.animation import FuncAnimation, FFMpegWriter, PillowWriter
|
21 |
+
import requests
|
22 |
+
import zipfile
|
23 |
+
|
24 |
+
|
25 |
+
warnings.filterwarnings("ignore")
|
26 |
+
|
27 |
+
|
28 |
+
def get_svg_content(svg_path):
|
29 |
+
with open(svg_path, "r") as file:
|
30 |
+
return file.read()
|
31 |
+
|
32 |
+
|
33 |
+
def download_file(url, filename):
|
34 |
+
response = requests.get(url)
|
35 |
+
with open(filename, "wb") as f:
|
36 |
+
f.write(response.content)
|
37 |
+
|
38 |
+
|
39 |
+
def unzip_file(filename, extract_to="."):
|
40 |
+
with zipfile.ZipFile(filename, "r") as zip_ref:
|
41 |
+
zip_ref.extractall(extract_to)
|
42 |
+
|
43 |
+
|
44 |
+
def load_and_pad_img_dir(file_dir):
|
45 |
+
image_path = os.path.join(file_dir)
|
46 |
+
image = Image.open(image_path)
|
47 |
+
width, height = image.size
|
48 |
+
ratio = min(224 / width, 224 / height)
|
49 |
+
image = image.resize((int(width * ratio), int(height * ratio)))
|
50 |
+
width, height = image.size
|
51 |
+
if height < 224:
|
52 |
+
# If width is shorter than height pad top and bottom.
|
53 |
+
top_padding = (224 - height) // 2
|
54 |
+
bottom_padding = 224 - height - top_padding
|
55 |
+
padded_image = Image.new("RGB", (width, 224), (255, 255, 255))
|
56 |
+
padded_image.paste(image, (0, top_padding))
|
57 |
+
else:
|
58 |
+
# Otherwise pad left and right.
|
59 |
+
left_padding = (224 - width) // 2
|
60 |
+
right_padding = 224 - width - left_padding
|
61 |
+
padded_image = Image.new("RGB", (224, height), (255, 255, 255))
|
62 |
+
padded_image.paste(image, (left_padding, 0))
|
63 |
+
return padded_image
|
64 |
+
|
65 |
+
|
66 |
+
def plot_ink(ink, ax, lw=1.8, input_image=None, with_path=True, path_color="white"):
|
67 |
+
if input_image is not None:
|
68 |
+
img = copy.deepcopy(input_image)
|
69 |
+
enhancer = ImageEnhance.Brightness(img)
|
70 |
+
img = enhancer.enhance(0.45)
|
71 |
+
ax.imshow(img)
|
72 |
+
|
73 |
+
base_colors = plt.cm.get_cmap("rainbow", len(ink.strokes))
|
74 |
+
|
75 |
+
for i, stroke in enumerate(ink.strokes):
|
76 |
+
x, y = np.array(stroke.x), np.array(stroke.y)
|
77 |
+
|
78 |
+
base_color = base_colors(len(ink.strokes) - 1 - i)
|
79 |
+
hsv_color = colorsys.rgb_to_hsv(*base_color[:3])
|
80 |
+
|
81 |
+
darker_color = colorsys.hsv_to_rgb(
|
82 |
+
hsv_color[0], hsv_color[1], max(0, hsv_color[2] * 0.65)
|
83 |
+
)
|
84 |
+
colors = [
|
85 |
+
mcolors.to_rgba(darker_color, alpha=1 - (0.5 * j / len(x)))
|
86 |
+
for j in range(len(x))
|
87 |
+
]
|
88 |
+
|
89 |
+
points = np.array([x, y]).T.reshape(-1, 1, 2)
|
90 |
+
segments = np.concatenate([points[:-1], points[1:]], axis=1)
|
91 |
+
|
92 |
+
lc = LineCollection(segments, colors=colors, linewidth=lw)
|
93 |
+
if with_path:
|
94 |
+
lc.set_path_effects(
|
95 |
+
[withStroke(linewidth=lw * 1.25, foreground=path_color)]
|
96 |
+
)
|
97 |
+
ax.add_collection(lc)
|
98 |
+
|
99 |
+
ax.set_xlim(0, 224)
|
100 |
+
ax.set_ylim(0, 224)
|
101 |
+
ax.invert_yaxis()
|
102 |
+
|
103 |
+
|
104 |
+
def plot_ink_to_video(
|
105 |
+
ink, output_name, lw=1.8, input_image=None, path_color="white", fps=30
|
106 |
+
):
|
107 |
+
fig, ax = plt.subplots(figsize=(4, 4), dpi=150)
|
108 |
+
|
109 |
+
if input_image is not None:
|
110 |
+
img = copy.deepcopy(input_image)
|
111 |
+
enhancer = ImageEnhance.Brightness(img)
|
112 |
+
img = enhancer.enhance(0.45)
|
113 |
+
ax.imshow(img)
|
114 |
+
|
115 |
+
ax.set_xlim(0, 224)
|
116 |
+
ax.set_ylim(0, 224)
|
117 |
+
ax.invert_yaxis()
|
118 |
+
ax.axis("off")
|
119 |
+
|
120 |
+
base_colors = plt.cm.get_cmap("rainbow", len(ink.strokes))
|
121 |
+
all_points = sum([len(stroke.x) for stroke in ink.strokes], 0)
|
122 |
+
|
123 |
+
def update(frame):
|
124 |
+
ax.clear()
|
125 |
+
if input_image is not None:
|
126 |
+
ax.imshow(img)
|
127 |
+
ax.set_xlim(0, 224)
|
128 |
+
ax.set_ylim(0, 224)
|
129 |
+
ax.invert_yaxis()
|
130 |
+
ax.axis("off")
|
131 |
+
|
132 |
+
points_drawn = 0
|
133 |
+
for stroke_index, stroke in enumerate(ink.strokes):
|
134 |
+
x, y = np.array(stroke.x), np.array(stroke.y)
|
135 |
+
points = np.array([x, y]).T.reshape(-1, 1, 2)
|
136 |
+
segments = np.concatenate([points[:-1], points[1:]], axis=1)
|
137 |
+
|
138 |
+
base_color = base_colors(len(ink.strokes) - 1 - stroke_index)
|
139 |
+
hsv_color = colorsys.rgb_to_hsv(*base_color[:3])
|
140 |
+
darker_color = colorsys.hsv_to_rgb(
|
141 |
+
hsv_color[0], hsv_color[1], max(0, hsv_color[2] * 0.65)
|
142 |
+
)
|
143 |
+
visible_segments = (
|
144 |
+
segments[: frame - points_drawn]
|
145 |
+
if frame - points_drawn < len(segments)
|
146 |
+
else segments
|
147 |
+
)
|
148 |
+
colors = [
|
149 |
+
mcolors.to_rgba(
|
150 |
+
darker_color, alpha=1 - (0.5 * j / len(visible_segments))
|
151 |
+
)
|
152 |
+
for j in range(len(visible_segments))
|
153 |
+
]
|
154 |
+
|
155 |
+
if len(visible_segments) > 0:
|
156 |
+
lc = LineCollection(visible_segments, colors=colors, linewidth=lw)
|
157 |
+
lc.set_path_effects(
|
158 |
+
[withStroke(linewidth=lw * 1.25, foreground=path_color)]
|
159 |
+
)
|
160 |
+
ax.add_collection(lc)
|
161 |
+
|
162 |
+
points_drawn += len(segments)
|
163 |
+
if points_drawn >= frame:
|
164 |
+
break
|
165 |
+
|
166 |
+
ani = FuncAnimation(fig, update, frames=all_points + 1, blit=False)
|
167 |
+
Writer = FFMpegWriter(fps=fps)
|
168 |
+
ani.save(output_name, writer=Writer)
|
169 |
+
plt.close(fig)
|
170 |
+
|
171 |
+
|
172 |
+
class Stroke:
|
173 |
+
def __init__(self, list_of_coordinates=None) -> None:
|
174 |
+
self.x = []
|
175 |
+
self.y = []
|
176 |
+
if list_of_coordinates:
|
177 |
+
for point in list_of_coordinates:
|
178 |
+
self.x.append(point[0])
|
179 |
+
self.y.append(point[1])
|
180 |
+
|
181 |
+
def __len__(self):
|
182 |
+
return len(self.x)
|
183 |
+
|
184 |
+
def __getitem__(self, index):
|
185 |
+
return (self.x[index], self.y[index])
|
186 |
+
|
187 |
+
|
188 |
+
class Ink:
|
189 |
+
def __init__(self, list_of_strokes=None) -> None:
|
190 |
+
self.strokes = []
|
191 |
+
if list_of_strokes:
|
192 |
+
self.strokes = list_of_strokes
|
193 |
+
|
194 |
+
def __len__(self):
|
195 |
+
return len(self.strokes)
|
196 |
+
|
197 |
+
def __getitem__(self, index):
|
198 |
+
return self.strokes[index]
|
199 |
+
|
200 |
+
|
201 |
+
def inkml_to_ink(inkml_file):
|
202 |
+
"""Convert inkml file to Ink"""
|
203 |
+
tree = ET.parse(inkml_file)
|
204 |
+
root = tree.getroot()
|
205 |
+
|
206 |
+
inkml_namespace = {"inkml": "http://www.w3.org/2003/InkML"}
|
207 |
+
|
208 |
+
strokes = []
|
209 |
+
|
210 |
+
for trace in root.findall("inkml:trace", inkml_namespace):
|
211 |
+
points = trace.text.strip().split()
|
212 |
+
stroke_points = []
|
213 |
+
|
214 |
+
for point in points:
|
215 |
+
x, y = point.split(",")
|
216 |
+
stroke_points.append((float(x), float(y)))
|
217 |
+
strokes.append(Stroke(stroke_points))
|
218 |
+
return Ink(strokes)
|
219 |
+
|
220 |
+
|
221 |
+
def parse_inkml_annotations(inkml_file):
|
222 |
+
tree = ET.parse(inkml_file)
|
223 |
+
root = tree.getroot()
|
224 |
+
|
225 |
+
annotations = root.findall(".//{http://www.w3.org/2003/InkML}annotation")
|
226 |
+
|
227 |
+
annotation_dict = {}
|
228 |
+
|
229 |
+
for annotation in annotations:
|
230 |
+
annotation_type = annotation.get("type")
|
231 |
+
annotation_text = annotation.text
|
232 |
+
|
233 |
+
annotation_dict[annotation_type] = annotation_text
|
234 |
+
|
235 |
+
return annotation_dict
|