Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,127 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import gradio as gr
|
3 |
+
import PIL.Image
|
4 |
+
import torch
|
5 |
+
from transformers import PaliGemmaForConditionalGeneration, PaliGemmaProcessor
|
6 |
+
|
7 |
+
# Model and Processor Setup
|
8 |
+
model_id = "gv-hf/paligemma2-3b-mix-448"
|
9 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
10 |
+
HF_KEY = os.getenv("HF_KEY")
|
11 |
+
if not HF_KEY:
|
12 |
+
raise ValueError("Please set the HF_KEY environment variable with your Hugging Face API token")
|
13 |
+
|
14 |
+
model = PaliGemmaForConditionalGeneration.from_pretrained(
|
15 |
+
model_id,
|
16 |
+
token=HF_KEY,
|
17 |
+
trust_remote_code=True
|
18 |
+
).eval().to(device)
|
19 |
+
|
20 |
+
processor = PaliGemmaProcessor.from_pretrained(
|
21 |
+
model_id,
|
22 |
+
token=HF_KEY,
|
23 |
+
trust_remote_code=True
|
24 |
+
)
|
25 |
+
|
26 |
+
# Inference Function
|
27 |
+
def infer(image: PIL.Image.Image, text: str, max_new_tokens: int) -> str:
|
28 |
+
inputs = processor(text=text, images=image, return_tensors="pt").to(device)
|
29 |
+
with torch.inference_mode():
|
30 |
+
generated_ids = model.generate(
|
31 |
+
**inputs,
|
32 |
+
max_new_tokens=max_new_tokens,
|
33 |
+
do_sample=False
|
34 |
+
)
|
35 |
+
result = processor.batch_decode(generated_ids, skip_special_tokens=True)
|
36 |
+
return result[0][len(text):].lstrip("\n")
|
37 |
+
|
38 |
+
# Image Captioning
|
39 |
+
def generate_caption(image: PIL.Image.Image) -> str:
|
40 |
+
return infer(image, "caption", max_new_tokens=50)
|
41 |
+
|
42 |
+
# Object Detection
|
43 |
+
def detect_objects(image: PIL.Image.Image) -> str:
|
44 |
+
return infer(image, "detect objects", max_new_tokens=200)
|
45 |
+
|
46 |
+
# Visual Question Answering (VQA)
|
47 |
+
def vqa(image: PIL.Image.Image, question: str) -> str:
|
48 |
+
return infer(image, f"Q: {question} A:", max_new_tokens=50)
|
49 |
+
|
50 |
+
# Custom CSS for Styling
|
51 |
+
custom_css = """
|
52 |
+
.gradio-container {
|
53 |
+
font-family: 'Arial', sans-serif;
|
54 |
+
}
|
55 |
+
.upload-button {
|
56 |
+
background-color: #4285f4;
|
57 |
+
color: white;
|
58 |
+
border-radius: 5px;
|
59 |
+
padding: 10px 20px;
|
60 |
+
}
|
61 |
+
.output-text {
|
62 |
+
font-size: 18px;
|
63 |
+
font-weight: bold;
|
64 |
+
}
|
65 |
+
"""
|
66 |
+
|
67 |
+
# Gradio App
|
68 |
+
with gr.Blocks(css=custom_css) as demo:
|
69 |
+
gr.Markdown("# PaliGemma Multi-Modal App")
|
70 |
+
gr.Markdown("Upload an image and explore its features using the PaliGemma model!")
|
71 |
+
|
72 |
+
with gr.Tabs():
|
73 |
+
# Tab 1: Image Captioning
|
74 |
+
with gr.Tab("Image Captioning"):
|
75 |
+
with gr.Row():
|
76 |
+
with gr.Column():
|
77 |
+
caption_image = gr.Image(type="pil", label="Upload Image", width=512, height=512)
|
78 |
+
caption_btn = gr.Button("Generate Caption", elem_classes="upload-button")
|
79 |
+
with gr.Column():
|
80 |
+
caption_output = gr.Text(label="Generated Caption", elem_classes="output-text")
|
81 |
+
caption_btn.click(fn=generate_caption, inputs=[caption_image], outputs=[caption_output])
|
82 |
+
|
83 |
+
# Tab 2: Object Detection
|
84 |
+
with gr.Tab("Object Detection"):
|
85 |
+
with gr.Row():
|
86 |
+
with gr.Column():
|
87 |
+
detect_image = gr.Image(type="pil", label="Upload Image", width=512, height=512)
|
88 |
+
detect_btn = gr.Button("Detect Objects", elem_classes="upload-button")
|
89 |
+
with gr.Column():
|
90 |
+
detect_output = gr.Text(label="Detected Objects", elem_classes="output-text")
|
91 |
+
detect_btn.click(fn=detect_objects, inputs=[detect_image], outputs=[detect_output])
|
92 |
+
|
93 |
+
# Tab 3: Visual Question Answering (VQA)
|
94 |
+
with gr.Tab("Visual Question Answering"):
|
95 |
+
with gr.Row():
|
96 |
+
with gr.Column():
|
97 |
+
vqa_image = gr.Image(type="pil", label="Upload Image", width=512, height=512)
|
98 |
+
vqa_question = gr.Text(label="Ask a Question", placeholder="What is in the image?")
|
99 |
+
vqa_btn = gr.Button("Ask", elem_classes="upload-button")
|
100 |
+
with gr.Column():
|
101 |
+
vqa_output = gr.Text(label="Answer", elem_classes="output-text")
|
102 |
+
vqa_btn.click(fn=vqa, inputs=[vqa_image, vqa_question], outputs=[vqa_output])
|
103 |
+
|
104 |
+
# Tab 4: Text Generation (Original Feature)
|
105 |
+
with gr.Tab("Text Generation"):
|
106 |
+
with gr.Row():
|
107 |
+
with gr.Column():
|
108 |
+
text_image = gr.Image(type="pil", label="Upload Image", width=512, height=512)
|
109 |
+
text_input = gr.Text(label="Input Text", placeholder="Describe the image...")
|
110 |
+
text_btn = gr.Button("Generate Text", elem_classes="upload-button")
|
111 |
+
with gr.Column():
|
112 |
+
text_output = gr.Text(label="Generated Text", elem_classes="output-text")
|
113 |
+
text_btn.click(fn=infer, inputs=[text_image, text_input, gr.Slider(10, 200, value=50)], outputs=[text_output])
|
114 |
+
|
115 |
+
# Image Upload/Download
|
116 |
+
with gr.Row():
|
117 |
+
upload_button = gr.UploadButton("Upload Image", file_types=["image"], elem_classes="upload-button")
|
118 |
+
download_button = gr.DownloadButton("Download Results", elem_classes="upload-button")
|
119 |
+
|
120 |
+
# Real-Time Updates
|
121 |
+
caption_image.change(fn=generate_caption, inputs=[caption_image], outputs=[caption_output], live=True)
|
122 |
+
detect_image.change(fn=detect_objects, inputs=[detect_image], outputs=[detect_output], live=True)
|
123 |
+
vqa_image.change(fn=lambda x: vqa(x, "What is in the image?"), inputs=[vqa_image], outputs=[vqa_output], live=True)
|
124 |
+
|
125 |
+
# Launch the App
|
126 |
+
if __name__ == "__main__":
|
127 |
+
demo.queue(max_size=10).launch(debug=True)
|