Update app.py
Browse files
app.py
CHANGED
|
@@ -6,257 +6,142 @@ import io
|
|
| 6 |
import json
|
| 7 |
from transformers import BlipProcessor, BlipForConditionalGeneration
|
| 8 |
import torch
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
|
| 10 |
# Initialize the BLIP model for image captioning
|
| 11 |
processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-base")
|
| 12 |
-
model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-base")
|
| 13 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
def generate_alt_text(image):
|
| 15 |
-
"""Generate alt text for a single image using BLIP model"""
|
| 16 |
try:
|
| 17 |
-
#
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
with torch.no_grad():
|
| 22 |
out = model.generate(
|
| 23 |
-
**inputs,
|
| 24 |
-
max_length=50,
|
| 25 |
-
min_length=5,
|
| 26 |
-
num_beams=3,
|
| 27 |
early_stopping=True,
|
| 28 |
-
no_repeat_ngram_size=2,
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
repetition_penalty=1.2 # Penalize repetition
|
| 33 |
)
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
# Clean up repetitive text
|
| 39 |
caption = clean_repetitive_text(caption)
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
elif caption.lower().startswith("an image of"):
|
| 45 |
-
caption = caption[11:].strip()
|
| 46 |
-
elif caption.lower().startswith("a photo of"):
|
| 47 |
-
caption = caption[10:].strip()
|
| 48 |
-
|
| 49 |
-
# Remove common prefixes and clean up
|
| 50 |
-
caption = caption.strip()
|
| 51 |
-
|
| 52 |
-
# Ensure it's not empty and capitalize first letter
|
| 53 |
-
if caption and len(caption) > 0:
|
| 54 |
-
caption = caption[0].upper() + caption[1:]
|
| 55 |
-
else:
|
| 56 |
-
caption = "Image content could not be determined"
|
| 57 |
-
|
| 58 |
-
# Final check for repetitive patterns
|
| 59 |
if is_repetitive(caption):
|
| 60 |
caption = "Image with visual content"
|
| 61 |
-
|
| 62 |
-
return caption
|
| 63 |
-
|
| 64 |
-
except Exception as e:
|
| 65 |
-
return f"Error generating alt text: {str(e)}"
|
| 66 |
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
|
| 71 |
-
# Remove patterns like "lan lan lan" or "a a a"
|
| 72 |
-
text = re.sub(r'\b(\w+)(\s+\1){2,}\b', r'\1', text)
|
| 73 |
-
|
| 74 |
-
# Remove excessive repetition of short words
|
| 75 |
-
text = re.sub(r'\b(\w{1,3})\s+(\1\s+){3,}', '', text)
|
| 76 |
-
|
| 77 |
-
# Clean up extra spaces
|
| 78 |
-
text = ' '.join(text.split())
|
| 79 |
-
|
| 80 |
-
return text.strip()
|
| 81 |
-
|
| 82 |
-
def is_repetitive(text):
|
| 83 |
-
"""Check if text is repetitive"""
|
| 84 |
-
words = text.split()
|
| 85 |
-
if len(words) < 3:
|
| 86 |
-
return False
|
| 87 |
-
|
| 88 |
-
# Check for repeated words
|
| 89 |
-
word_counts = {}
|
| 90 |
-
for word in words:
|
| 91 |
-
word_counts[word] = word_counts.get(word, 0) + 1
|
| 92 |
-
|
| 93 |
-
# If any word appears more than 3 times, it's likely repetitive
|
| 94 |
-
for count in word_counts.values():
|
| 95 |
-
if count > 3:
|
| 96 |
-
return True
|
| 97 |
-
|
| 98 |
-
return False
|
| 99 |
|
|
|
|
| 100 |
def process_images(files):
|
| 101 |
-
"""Process multiple uploaded images and generate alt text for each"""
|
| 102 |
-
if not files:
|
| 103 |
-
return "No images uploaded"
|
| 104 |
-
|
| 105 |
results = []
|
| 106 |
-
|
| 107 |
-
|
| 108 |
try:
|
| 109 |
-
|
| 110 |
-
|
| 111 |
-
|
| 112 |
-
|
| 113 |
-
|
| 114 |
-
|
| 115 |
-
|
| 116 |
-
result = {
|
| 117 |
-
"image_index": i + 1,
|
| 118 |
-
"filename": file.name.split('/')[-1],
|
| 119 |
-
"alt_text": alt_text,
|
| 120 |
-
"image": image
|
| 121 |
-
}
|
| 122 |
-
results.append(result)
|
| 123 |
-
|
| 124 |
except Exception as e:
|
| 125 |
results.append({
|
| 126 |
-
|
| 127 |
-
|
| 128 |
-
|
| 129 |
-
"image": None
|
| 130 |
})
|
| 131 |
-
|
| 132 |
return results
|
| 133 |
|
| 134 |
-
|
| 135 |
-
|
| 136 |
-
|
| 137 |
-
|
| 138 |
-
|
| 139 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 140 |
"""
|
| 141 |
-
|
| 142 |
-
for i, result in enumerate(results):
|
| 143 |
-
button_id = f"copy-btn-{i}"
|
| 144 |
html += f"""
|
| 145 |
-
<div class="result-container">
|
| 146 |
-
|
| 147 |
-
|
| 148 |
-
|
| 149 |
-
|
| 150 |
-
|
| 151 |
-
|
| 152 |
-
|
| 153 |
-
|
| 154 |
</div>
|
| 155 |
"""
|
| 156 |
-
|
| 157 |
html += "</div>"
|
| 158 |
return html
|
| 159 |
|
|
|
|
| 160 |
def main_process(files):
|
| 161 |
-
"""Main processing function for the Gradio interface"""
|
| 162 |
if not files:
|
| 163 |
return "<p>Please upload at least one image to generate alt text.</p>", []
|
| 164 |
-
|
| 165 |
-
# Process all uploaded images
|
| 166 |
results = process_images(files)
|
| 167 |
-
|
| 168 |
-
|
| 169 |
-
|
| 170 |
-
|
| 171 |
-
# Create gallery of images for display
|
| 172 |
-
gallery_images = []
|
| 173 |
-
for result in results:
|
| 174 |
-
if result['image']:
|
| 175 |
-
gallery_images.append(result['image'])
|
| 176 |
-
|
| 177 |
-
return html_output, gallery_images
|
| 178 |
|
| 179 |
-
#
|
| 180 |
-
|
|
|
|
| 181 |
gr.Markdown("""
|
| 182 |
# 🖼️ Alt Text Generator
|
| 183 |
-
|
| 184 |
-
Upload multiple images and generate descriptive alt text for each one.
|
| 185 |
-
The AI will analyze both visual content and any text present in the images.
|
| 186 |
-
|
| 187 |
-
**Features:**
|
| 188 |
-
- Support for multiple image uploads
|
| 189 |
-
- AI-powered alt text generation
|
| 190 |
-
- Copy-to-clipboard functionality for each generated alt text
|
| 191 |
-
- Preview of all uploaded images
|
| 192 |
""")
|
| 193 |
-
|
| 194 |
with gr.Row():
|
| 195 |
with gr.Column():
|
| 196 |
-
file_input = gr.Files(
|
| 197 |
-
|
| 198 |
-
file_count="multiple",
|
| 199 |
-
file_types=["image"],
|
| 200 |
-
height=200
|
| 201 |
-
)
|
| 202 |
-
|
| 203 |
-
process_btn = gr.Button(
|
| 204 |
-
"Generate Alt Text",
|
| 205 |
-
variant="primary",
|
| 206 |
-
size="lg"
|
| 207 |
-
)
|
| 208 |
-
|
| 209 |
with gr.Column():
|
| 210 |
-
|
| 211 |
-
|
| 212 |
-
|
| 213 |
-
|
| 214 |
-
columns=3,
|
| 215 |
-
rows=2,
|
| 216 |
-
height="auto"
|
| 217 |
-
)
|
| 218 |
-
|
| 219 |
-
with gr.Row():
|
| 220 |
-
output_html = gr.HTML(
|
| 221 |
-
label="Generated Alt Text Results",
|
| 222 |
-
show_label=True
|
| 223 |
-
)
|
| 224 |
-
|
| 225 |
-
# Event handlers
|
| 226 |
-
process_btn.click(
|
| 227 |
-
fn=main_process,
|
| 228 |
-
inputs=[file_input],
|
| 229 |
-
outputs=[output_html, image_gallery]
|
| 230 |
-
)
|
| 231 |
-
|
| 232 |
-
# Auto-process when files are uploaded
|
| 233 |
-
file_input.change(
|
| 234 |
-
fn=main_process,
|
| 235 |
-
inputs=[file_input],
|
| 236 |
-
outputs=[output_html, image_gallery]
|
| 237 |
-
)
|
| 238 |
-
|
| 239 |
-
gr.Markdown("""
|
| 240 |
-
### 📝 How to Use:
|
| 241 |
-
1. **Upload Images**: Click "Upload Images" and select one or more image files
|
| 242 |
-
2. **Generate Alt Text**: Click "Generate Alt Text" or wait for automatic processing
|
| 243 |
-
3. **Copy Alt Text**: Use the "Copy Alt Text" button next to each generated description
|
| 244 |
-
4. **Use the Alt Text**: Paste the copied text into your HTML `alt` attributes
|
| 245 |
-
|
| 246 |
-
### 🎯 About Alt Text:
|
| 247 |
-
Alt text (alternative text) is crucial for web accessibility. It helps screen readers describe images to visually impaired users and provides context when images fail to load.
|
| 248 |
-
|
| 249 |
-
### 🔧 Technical Details:
|
| 250 |
-
- Uses Salesforce's BLIP model for image captioning
|
| 251 |
-
- Processes both visual elements and text within images
|
| 252 |
-
- Generates human-readable descriptions suitable for alt attributes
|
| 253 |
-
""")
|
| 254 |
|
| 255 |
-
|
| 256 |
-
if __name__ == "__main__":
|
| 257 |
-
demo.launch(
|
| 258 |
-
server_name="0.0.0.0",
|
| 259 |
-
server_port=7860,
|
| 260 |
-
share=True,
|
| 261 |
-
debug=True
|
| 262 |
-
)
|
|
|
|
| 6 |
import json
|
| 7 |
from transformers import BlipProcessor, BlipForConditionalGeneration
|
| 8 |
import torch
|
| 9 |
+
import easyocr
|
| 10 |
+
|
| 11 |
+
# Initialize OCR reader (English)
|
| 12 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
| 13 |
+
ocr_reader = easyocr.Reader(['en'], gpu=(device=='cuda'))
|
| 14 |
|
| 15 |
# Initialize the BLIP model for image captioning
|
| 16 |
processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-base")
|
| 17 |
+
model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-base").to(device)
|
| 18 |
|
| 19 |
+
# Helper functions for cleaning captions
|
| 20 |
+
def clean_repetitive_text(text):
|
| 21 |
+
import re
|
| 22 |
+
text = re.sub(r'\b(\w+)(\s+\1){2,}\b', r'\1', text)
|
| 23 |
+
text = re.sub(r'\b(\w{1,3})\s+(\1\s+){3,}', '', text)
|
| 24 |
+
return ' '.join(text.split()).strip()
|
| 25 |
+
|
| 26 |
+
def is_repetitive(text):
|
| 27 |
+
words = text.split()
|
| 28 |
+
if len(words) < 3:
|
| 29 |
+
return False
|
| 30 |
+
from collections import Counter
|
| 31 |
+
return any(count > 3 for count in Counter(words).values())
|
| 32 |
+
|
| 33 |
+
# Generate alt text for a single image
|
| 34 |
def generate_alt_text(image):
|
|
|
|
| 35 |
try:
|
| 36 |
+
# OCR to extract visible text
|
| 37 |
+
buf = io.BytesIO()
|
| 38 |
+
image.save(buf, format='PNG')
|
| 39 |
+
buf.seek(0)
|
| 40 |
+
ocr_results = ocr_reader.readtext(buf, detail=0)
|
| 41 |
+
visible_text = ' '.join(ocr_results).strip() or '<no text detected>'
|
| 42 |
+
|
| 43 |
+
# Generate caption
|
| 44 |
+
inputs = processor(images=image, return_tensors="pt").to(device)
|
| 45 |
with torch.no_grad():
|
| 46 |
out = model.generate(
|
| 47 |
+
**inputs,
|
| 48 |
+
max_length=50,
|
| 49 |
+
min_length=5,
|
| 50 |
+
num_beams=3,
|
| 51 |
early_stopping=True,
|
| 52 |
+
no_repeat_ngram_size=2,
|
| 53 |
+
temperature=0.7,
|
| 54 |
+
top_p=0.9,
|
| 55 |
+
repetition_penalty=1.2
|
|
|
|
| 56 |
)
|
| 57 |
+
caption = processor.decode(out[0], skip_special_tokens=True).strip()
|
| 58 |
+
|
| 59 |
+
# Clean caption
|
|
|
|
|
|
|
| 60 |
caption = clean_repetitive_text(caption)
|
| 61 |
+
for prefix in ["a picture of", "an image of", "a photo of"]:
|
| 62 |
+
if caption.lower().startswith(prefix):
|
| 63 |
+
caption = caption[len(prefix):].strip()
|
| 64 |
+
caption = caption.capitalize() if caption else "Image content could not be determined"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 65 |
if is_repetitive(caption):
|
| 66 |
caption = "Image with visual content"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 67 |
|
| 68 |
+
# Combine with OCR text
|
| 69 |
+
return f"{caption} (Text on image: {visible_text})"
|
| 70 |
+
except Exception as e:
|
| 71 |
+
return f"Error generating alt text: {e}"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 72 |
|
| 73 |
+
# Process multiple files
|
| 74 |
def process_images(files):
|
|
|
|
|
|
|
|
|
|
|
|
|
| 75 |
results = []
|
| 76 |
+
for i, f in enumerate(files):
|
| 77 |
+
path = f.name if hasattr(f, 'name') else None
|
| 78 |
try:
|
| 79 |
+
img = Image.open(path).convert('RGB') if path else Image.open(f).convert('RGB')
|
| 80 |
+
alt = generate_alt_text(img)
|
| 81 |
+
results.append({
|
| 82 |
+
'filename': path.split('/')[-1] if path else f'Image {i+1}',
|
| 83 |
+
'alt_text': alt,
|
| 84 |
+
'image': img
|
| 85 |
+
})
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 86 |
except Exception as e:
|
| 87 |
results.append({
|
| 88 |
+
'filename': path.split('/')[-1] if path else f'Image {i+1}',
|
| 89 |
+
'alt_text': f"Error processing image: {e}",
|
| 90 |
+
'image': None
|
|
|
|
| 91 |
})
|
|
|
|
| 92 |
return results
|
| 93 |
|
| 94 |
+
# Create HTML to display results with copy buttons
|
| 95 |
+
def create_result_html(results):
|
| 96 |
+
html = """
|
| 97 |
+
<style>
|
| 98 |
+
.result-container { margin-bottom: 1rem; }
|
| 99 |
+
.result-container .filename { font-weight: bold; color: #333; margin-bottom: 5px; }
|
| 100 |
+
.result-container .alt-text-container { margin-bottom: 5px; }
|
| 101 |
+
.copy-button { padding: 0.3em 0.6em; margin-top: 5px; }
|
| 102 |
+
</style>
|
| 103 |
+
<div>
|
| 104 |
"""
|
| 105 |
+
for i, r in enumerate(results):
|
|
|
|
|
|
|
| 106 |
html += f"""
|
| 107 |
+
<div class=\"result-container\">
|
| 108 |
+
<div class=\"filename\">📁 {r['filename']}</div>
|
| 109 |
+
<div class=\"alt-text-container\">
|
| 110 |
+
<strong>Alt Text:</strong><br>
|
| 111 |
+
<span id=\"alt-text-{i}\">{r['alt_text']}</span>
|
| 112 |
+
</div>
|
| 113 |
+
<button class=\"copy-button\" onclick=\"navigator.clipboard.writeText(document.getElementById('alt-text-{i}').innerText)\">
|
| 114 |
+
📋 Copy Alt Text
|
| 115 |
+
</button>
|
| 116 |
</div>
|
| 117 |
"""
|
|
|
|
| 118 |
html += "</div>"
|
| 119 |
return html
|
| 120 |
|
| 121 |
+
# Main processing function for Gradio
|
| 122 |
def main_process(files):
|
|
|
|
| 123 |
if not files:
|
| 124 |
return "<p>Please upload at least one image to generate alt text.</p>", []
|
|
|
|
|
|
|
| 125 |
results = process_images(files)
|
| 126 |
+
html = create_result_html(results)
|
| 127 |
+
gallery = [r['image'] for r in results if r['image']]
|
| 128 |
+
return html, gallery
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 129 |
|
| 130 |
+
# Build the Gradio app
|
| 131 |
+
demo = gr.Blocks()
|
| 132 |
+
with demo:
|
| 133 |
gr.Markdown("""
|
| 134 |
# 🖼️ Alt Text Generator
|
| 135 |
+
Upload multiple images and generate descriptive alt text for each one. The AI analyzes visual content and any text present.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 136 |
""")
|
|
|
|
| 137 |
with gr.Row():
|
| 138 |
with gr.Column():
|
| 139 |
+
file_input = gr.Files(label="Upload Images", file_count="multiple", file_types=["image"], height=200)
|
| 140 |
+
process_btn = gr.Button("Generate Alt Text", variant="primary", size="lg")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 141 |
with gr.Column():
|
| 142 |
+
gallery = gr.Gallery(label="Uploaded Images", show_label=True, columns=3)
|
| 143 |
+
output_html = gr.HTML(label="Generated Alt Text Results")
|
| 144 |
+
process_btn.click(fn=main_process, inputs=file_input, outputs=[output_html, gallery])
|
| 145 |
+
file_input.change(fn=main_process, inputs=file_input, outputs=[output_html, gallery])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 146 |
|
| 147 |
+
demo.launch(server_name="0.0.0.0", server_port=7860, share=True, debug=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|