Upload visual_qa (2).py
Browse files- visual_qa (2).py +192 -0
visual_qa (2).py
ADDED
@@ -0,0 +1,192 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from PIL import Image
|
2 |
+
import base64
|
3 |
+
from io import BytesIO
|
4 |
+
import json
|
5 |
+
import os
|
6 |
+
import requests
|
7 |
+
from typing import Optional
|
8 |
+
from huggingface_hub import InferenceClient
|
9 |
+
from transformers import AutoProcessor, Tool
|
10 |
+
import uuid
|
11 |
+
import mimetypes
|
12 |
+
##from dotenv import load_dotenv
|
13 |
+
|
14 |
+
##load_dotenv(override=True)
|
15 |
+
|
16 |
+
idefics_processor = AutoProcessor.from_pretrained("HuggingFaceM4/idefics2-8b")
|
17 |
+
|
18 |
+
def process_images_and_text(image_path, query, client):
|
19 |
+
messages = [
|
20 |
+
{
|
21 |
+
"role": "user", "content": [
|
22 |
+
{"type": "image"},
|
23 |
+
{"type": "text", "text": query},
|
24 |
+
]
|
25 |
+
},
|
26 |
+
]
|
27 |
+
|
28 |
+
prompt_with_template = idefics_processor.apply_chat_template(messages, add_generation_prompt=True)
|
29 |
+
|
30 |
+
# load images from local directory
|
31 |
+
|
32 |
+
# encode images to strings which can be sent to the endpoint
|
33 |
+
def encode_local_image(image_path):
|
34 |
+
# load image
|
35 |
+
image = Image.open(image_path).convert('RGB')
|
36 |
+
|
37 |
+
# Convert the image to a base64 string
|
38 |
+
buffer = BytesIO()
|
39 |
+
image.save(buffer, format="JPEG") # Use the appropriate format (e.g., JPEG, PNG)
|
40 |
+
base64_image = base64.b64encode(buffer.getvalue()).decode('utf-8')
|
41 |
+
|
42 |
+
# add string formatting required by the endpoint
|
43 |
+
image_string = f"data:image/jpeg;base64,{base64_image}"
|
44 |
+
|
45 |
+
return image_string
|
46 |
+
|
47 |
+
|
48 |
+
image_string = encode_local_image(image_path)
|
49 |
+
prompt_with_images = prompt_with_template.replace("<image>", "![]({}) ").format(image_string)
|
50 |
+
|
51 |
+
|
52 |
+
payload = {
|
53 |
+
"inputs": prompt_with_images,
|
54 |
+
"parameters": {
|
55 |
+
"return_full_text": False,
|
56 |
+
"max_new_tokens": 200,
|
57 |
+
}
|
58 |
+
}
|
59 |
+
|
60 |
+
return json.loads(client.post(json=payload).decode())[0]
|
61 |
+
|
62 |
+
# Function to encode the image
|
63 |
+
def encode_image(image_path):
|
64 |
+
if image_path.startswith("http"):
|
65 |
+
user_agent = "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/119.0.0.0 Safari/537.36 Edg/119.0.0.0"
|
66 |
+
request_kwargs = {
|
67 |
+
"headers": {"User-Agent": user_agent},
|
68 |
+
"stream": True,
|
69 |
+
}
|
70 |
+
|
71 |
+
# Send a HTTP request to the URL
|
72 |
+
response = requests.get(image_path, **request_kwargs)
|
73 |
+
response.raise_for_status()
|
74 |
+
content_type = response.headers.get("content-type", "")
|
75 |
+
|
76 |
+
extension = mimetypes.guess_extension(content_type)
|
77 |
+
if extension is None:
|
78 |
+
extension = ".download"
|
79 |
+
|
80 |
+
fname = str(uuid.uuid4()) + extension
|
81 |
+
download_path = os.path.abspath(os.path.join("downloads", fname))
|
82 |
+
|
83 |
+
with open(download_path, "wb") as fh:
|
84 |
+
for chunk in response.iter_content(chunk_size=512):
|
85 |
+
fh.write(chunk)
|
86 |
+
|
87 |
+
image_path = download_path
|
88 |
+
|
89 |
+
with open(image_path, "rb") as image_file:
|
90 |
+
return base64.b64encode(image_file.read()).decode('utf-8')
|
91 |
+
|
92 |
+
headers = {
|
93 |
+
"Content-Type": "application/json",
|
94 |
+
"Authorization": f"Bearer {os.getenv('OPENAI_API_KEY')}"
|
95 |
+
}
|
96 |
+
|
97 |
+
|
98 |
+
def resize_image(image_path):
|
99 |
+
img = Image.open(image_path)
|
100 |
+
width, height = img.size
|
101 |
+
img = img.resize((int(width / 2), int(height / 2)))
|
102 |
+
new_image_path = f"resized_{image_path}"
|
103 |
+
img.save(new_image_path)
|
104 |
+
return new_image_path
|
105 |
+
|
106 |
+
|
107 |
+
class VisualQATool(Tool):
|
108 |
+
name = "visualizer"
|
109 |
+
description = "A tool that can answer questions about attached images."
|
110 |
+
inputs = {
|
111 |
+
"question": {"description": "the question to answer", "type": "text"},
|
112 |
+
"image_path": {
|
113 |
+
"description": "The path to the image on which to answer the question",
|
114 |
+
"type": "text",
|
115 |
+
},
|
116 |
+
}
|
117 |
+
output_type = "text"
|
118 |
+
|
119 |
+
client = InferenceClient("HuggingFaceM4/idefics2-8b-chatty")
|
120 |
+
|
121 |
+
def forward(self, image_path: str, question: Optional[str] = None) -> str:
|
122 |
+
add_note = False
|
123 |
+
if not question:
|
124 |
+
add_note = True
|
125 |
+
question = "Please write a detailed caption for this image."
|
126 |
+
try:
|
127 |
+
output = process_images_and_text(image_path, question, self.client)
|
128 |
+
except Exception as e:
|
129 |
+
print(e)
|
130 |
+
if "Payload Too Large" in str(e):
|
131 |
+
new_image_path = resize_image(image_path)
|
132 |
+
output = process_images_and_text(new_image_path, question, self.client)
|
133 |
+
|
134 |
+
if add_note:
|
135 |
+
output = f"You did not provide a particular question, so here is a detailed caption for the image: {output}"
|
136 |
+
|
137 |
+
return output
|
138 |
+
|
139 |
+
class VisualQAGPT4Tool(Tool):
|
140 |
+
name = "visualizer"
|
141 |
+
description = "A tool that can answer questions about attached images."
|
142 |
+
inputs = {
|
143 |
+
"question": {"description": "the question to answer", "type": "text"},
|
144 |
+
"image_path": {
|
145 |
+
"description": "The path to the image on which to answer the question. This should be a local path to downloaded image.",
|
146 |
+
"type": "text",
|
147 |
+
},
|
148 |
+
}
|
149 |
+
output_type = "text"
|
150 |
+
|
151 |
+
def forward(self, image_path: str, question: Optional[str] = None) -> str:
|
152 |
+
add_note = False
|
153 |
+
if not question:
|
154 |
+
add_note = True
|
155 |
+
question = "Please write a detailed caption for this image."
|
156 |
+
if not isinstance(image_path, str):
|
157 |
+
raise Exception("You should provide only one string as argument to this tool!")
|
158 |
+
|
159 |
+
base64_image = encode_image(image_path)
|
160 |
+
|
161 |
+
payload = {
|
162 |
+
"model": "gpt-4o",
|
163 |
+
"messages": [
|
164 |
+
{
|
165 |
+
"role": "user",
|
166 |
+
"content": [
|
167 |
+
{
|
168 |
+
"type": "text",
|
169 |
+
"text": question
|
170 |
+
},
|
171 |
+
{
|
172 |
+
"type": "image_url",
|
173 |
+
"image_url": {
|
174 |
+
"url": f"data:image/jpeg;base64,{base64_image}"
|
175 |
+
}
|
176 |
+
}
|
177 |
+
]
|
178 |
+
}
|
179 |
+
],
|
180 |
+
"max_tokens": 500
|
181 |
+
}
|
182 |
+
response = requests.post("https://api.openai.com/v1/chat/completions", headers=headers, json=payload)
|
183 |
+
try:
|
184 |
+
output = response.json()['choices'][0]['message']['content']
|
185 |
+
except Exception:
|
186 |
+
raise Exception(f"Response format unexpected: {response.json()}")
|
187 |
+
|
188 |
+
if add_note:
|
189 |
+
output = f"You did not provide a particular question, so here is a detailed caption for the image: {output}"
|
190 |
+
|
191 |
+
return output
|
192 |
+
|