Spaces:
Sleeping
Sleeping
Commit
·
d0d6669
1
Parent(s):
00b05e0
Update to Get Donut Results
Browse files
app.py
CHANGED
@@ -1,20 +1,34 @@
|
|
1 |
import io
|
2 |
import requests
|
3 |
import gradio as gr
|
4 |
-
# from transformers import AutoModel, AutoTokenizer
|
5 |
from huggingface_hub import list_models
|
6 |
from datasets import load_dataset
|
7 |
from typing import List
|
8 |
from PIL import Image
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
9 |
|
10 |
|
11 |
def get_image_names(dataset):
|
12 |
return [str(i) for i in range(len(dataset))]
|
13 |
|
|
|
14 |
def get_image_from_dataset(index):
|
15 |
image_data = dataset[int(index)]["image"]
|
16 |
return image_data
|
17 |
|
|
|
18 |
def process_image(image=None, dataset_image_index=None):
|
19 |
if dataset_image_index:
|
20 |
image = get_image_from_dataset(dataset_image_index)
|
@@ -22,19 +36,20 @@ def process_image(image=None, dataset_image_index=None):
|
|
22 |
return image
|
23 |
|
24 |
|
25 |
-
|
26 |
def create_interface(tag, image_indices):
|
27 |
-
"""
|
28 |
iface = gr.Interface(
|
29 |
fn=process_image,
|
30 |
inputs=[
|
31 |
gr.Dropdown(choices=get_collection_models(tag), label="Select Model"),
|
32 |
gr.Image(type="pil", label="Upload Image"),
|
33 |
-
gr.Dropdown(
|
|
|
|
|
34 |
],
|
35 |
outputs=gr.Image(label="Output Image"),
|
36 |
title="Saliency Visualization",
|
37 |
-
description="Upload your image or select one from the MERIT Dataset test-set."
|
38 |
)
|
39 |
return iface
|
40 |
|
@@ -50,19 +65,102 @@ def get_collection_models(tag: str) -> List[str]:
|
|
50 |
|
51 |
return model_names
|
52 |
|
|
|
53 |
def load_model(model_name: str):
|
54 |
"""Load a model from Hugging Face Hub."""
|
55 |
model = AutoModel.from_pretrained(model_name)
|
56 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
57 |
return model, tokenizer
|
58 |
|
59 |
-
# # Example processing function
|
60 |
-
# def process_input(text: str, model_name: str) -> str:
|
61 |
-
# model, tokenizer = load_model(model_name)
|
62 |
-
# inputs = tokenizer(text, return_tensors="pt")
|
63 |
-
# outputs = model(**inputs)
|
64 |
-
# return f"Processed output with {model_name}"
|
65 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
66 |
|
67 |
dataset_name = "de-Rodrigo/merit"
|
68 |
dataset = load_dataset(dataset_name, name="en-digital-seq", split="train", num_proc=8)
|
|
|
1 |
import io
|
2 |
import requests
|
3 |
import gradio as gr
|
|
|
4 |
from huggingface_hub import list_models
|
5 |
from datasets import load_dataset
|
6 |
from typing import List
|
7 |
from PIL import Image
|
8 |
+
import torch
|
9 |
+
from transformers import DonutProcessor, VisionEncoderDecoderModel
|
10 |
+
import json
|
11 |
+
import re
|
12 |
+
import logging
|
13 |
+
|
14 |
+
# Logging configuration
|
15 |
+
logging.basicConfig(level=logging.INFO)
|
16 |
+
logger = logging.getLogger(__name__)
|
17 |
+
|
18 |
+
# Global variables for Donut model and processor
|
19 |
+
donut_model = None
|
20 |
+
donut_processor = None
|
21 |
|
22 |
|
23 |
def get_image_names(dataset):
|
24 |
return [str(i) for i in range(len(dataset))]
|
25 |
|
26 |
+
|
27 |
def get_image_from_dataset(index):
|
28 |
image_data = dataset[int(index)]["image"]
|
29 |
return image_data
|
30 |
|
31 |
+
|
32 |
def process_image(image=None, dataset_image_index=None):
|
33 |
if dataset_image_index:
|
34 |
image = get_image_from_dataset(dataset_image_index)
|
|
|
36 |
return image
|
37 |
|
38 |
|
|
|
39 |
def create_interface(tag, image_indices):
|
40 |
+
"""Create Gradio interface"""
|
41 |
iface = gr.Interface(
|
42 |
fn=process_image,
|
43 |
inputs=[
|
44 |
gr.Dropdown(choices=get_collection_models(tag), label="Select Model"),
|
45 |
gr.Image(type="pil", label="Upload Image"),
|
46 |
+
gr.Dropdown(
|
47 |
+
choices=image_indices, label="Select one from MERIT Dataset test-set"
|
48 |
+
),
|
49 |
],
|
50 |
outputs=gr.Image(label="Output Image"),
|
51 |
title="Saliency Visualization",
|
52 |
+
description="Upload your image or select one from the MERIT Dataset test-set.",
|
53 |
)
|
54 |
return iface
|
55 |
|
|
|
65 |
|
66 |
return model_names
|
67 |
|
68 |
+
|
69 |
def load_model(model_name: str):
|
70 |
"""Load a model from Hugging Face Hub."""
|
71 |
model = AutoModel.from_pretrained(model_name)
|
72 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
73 |
return model, tokenizer
|
74 |
|
|
|
|
|
|
|
|
|
|
|
|
|
75 |
|
76 |
+
def get_donut():
|
77 |
+
global donut_model, donut_processor
|
78 |
+
if donut_model is None or donut_processor is None:
|
79 |
+
try:
|
80 |
+
donut_model = VisionEncoderDecoderModel.from_pretrained(
|
81 |
+
"de-Rodrigo/donut-merit"
|
82 |
+
)
|
83 |
+
donut_processor = DonutProcessor.from_pretrained("de-Rodrigo/donut-merit")
|
84 |
+
if torch.cuda.is_available():
|
85 |
+
donut_model = donut_model.to("cuda")
|
86 |
+
logger.info("Donut model loaded successfully")
|
87 |
+
except Exception as e:
|
88 |
+
logger.error(f"Error loading Donut model: {str(e)}")
|
89 |
+
raise
|
90 |
+
return donut_model, donut_processor
|
91 |
+
|
92 |
+
|
93 |
+
def process_image_donut(model, processor, image):
|
94 |
+
try:
|
95 |
+
if not isinstance(image, Image.Image):
|
96 |
+
image = Image.fromarray(image)
|
97 |
+
|
98 |
+
pixel_values = processor(image, return_tensors="pt").pixel_values
|
99 |
+
if torch.cuda.is_available():
|
100 |
+
pixel_values = pixel_values.to("cuda")
|
101 |
+
|
102 |
+
task_prompt = "<s_cord-v2>"
|
103 |
+
decoder_input_ids = processor.tokenizer(
|
104 |
+
task_prompt, add_special_tokens=False, return_tensors="pt"
|
105 |
+
)["input_ids"]
|
106 |
+
|
107 |
+
outputs = model.generate(
|
108 |
+
pixel_values,
|
109 |
+
decoder_input_ids=decoder_input_ids,
|
110 |
+
max_length=model.decoder.config.max_position_embeddings,
|
111 |
+
early_stopping=True,
|
112 |
+
pad_token_id=processor.tokenizer.pad_token_id,
|
113 |
+
eos_token_id=processor.tokenizer.eos_token_id,
|
114 |
+
use_cache=True,
|
115 |
+
num_beams=1,
|
116 |
+
bad_words_ids=[[processor.tokenizer.unk_token_id]],
|
117 |
+
return_dict_in_generate=True,
|
118 |
+
)
|
119 |
+
|
120 |
+
sequence = processor.batch_decode(outputs.sequences)[0]
|
121 |
+
sequence = sequence.replace(processor.tokenizer.eos_token, "").replace(
|
122 |
+
processor.tokenizer.pad_token, ""
|
123 |
+
)
|
124 |
+
sequence = re.sub(r"<.*?>", "", sequence, count=1).strip()
|
125 |
+
|
126 |
+
result = processor.token2json(sequence)
|
127 |
+
return json.dumps(result, indent=2)
|
128 |
+
except Exception as e:
|
129 |
+
logger.error(f"Error processing image with Donut: {str(e)}")
|
130 |
+
return f"Error: {str(e)}"
|
131 |
+
|
132 |
+
|
133 |
+
def process_image(model_name, image=None, dataset_image_index=None):
|
134 |
+
if dataset_image_index is not None:
|
135 |
+
image = get_image_from_dataset(dataset_image_index)
|
136 |
+
|
137 |
+
if model_name == "de-Rodrigo/donut-merit":
|
138 |
+
model, processor = get_donut()
|
139 |
+
result = process_image_donut(model, processor, image)
|
140 |
+
else:
|
141 |
+
# Here you should implement processing for other models
|
142 |
+
result = f"Processing for model {model_name} not implemented"
|
143 |
+
|
144 |
+
return image, result
|
145 |
+
|
146 |
+
|
147 |
+
if __name__ == "__main__":
|
148 |
+
models = get_collection_models("saliency")
|
149 |
+
models.append("de-Rodrigo/donut-merit")
|
150 |
+
|
151 |
+
demo = gr.Interface(
|
152 |
+
fn=process_image,
|
153 |
+
inputs=[
|
154 |
+
gr.Dropdown(choices=models, label="Select Model"),
|
155 |
+
gr.Image(type="pil", label="Upload Image"),
|
156 |
+
gr.Slider(minimum=0, maximum=99, step=1, label="Dataset Image Index"),
|
157 |
+
],
|
158 |
+
outputs=[gr.Image(label="Processed Image"), gr.Textbox(label="Result")],
|
159 |
+
title="Document Understanding with Donut",
|
160 |
+
description="Upload an image or select one from the dataset to process with the selected model.",
|
161 |
+
)
|
162 |
+
|
163 |
+
demo.launch()
|
164 |
|
165 |
dataset_name = "de-Rodrigo/merit"
|
166 |
dataset = load_dataset(dataset_name, name="en-digital-seq", split="train", num_proc=8)
|