Workpam commited on
Commit
1221456
·
verified ·
1 Parent(s): bede987

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +64 -60
app.py CHANGED
@@ -17,27 +17,86 @@ def generate_alt_text(image):
17
  # Process the image
18
  inputs = processor(image, return_tensors="pt")
19
 
20
- # Generate caption
21
  with torch.no_grad():
22
- out = model.generate(**inputs, max_length=150, num_beams=5, early_stopping=True)
 
 
 
 
 
 
 
 
 
 
 
23
 
24
  # Decode the generated text
25
  caption = processor.decode(out[0], skip_special_tokens=True)
26
 
 
 
 
27
  # Clean up the caption to make it more suitable as alt text
28
  if caption.lower().startswith("a picture of"):
29
  caption = caption[12:].strip()
30
  elif caption.lower().startswith("an image of"):
31
  caption = caption[11:].strip()
 
 
 
 
 
32
 
33
- # Capitalize first letter
34
- caption = caption[0].upper() + caption[1:] if caption else "Image content could not be determined"
 
 
 
 
 
 
 
35
 
36
  return caption
37
 
38
  except Exception as e:
39
  return f"Error generating alt text: {str(e)}"
40
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
41
  def process_images(files):
42
  """Process multiple uploaded images and generate alt text for each"""
43
  if not files:
@@ -72,62 +131,7 @@ def process_images(files):
72
 
73
  return results
74
 
75
- def create_result_html(results):
76
- """Create HTML output with images, alt text, and copy buttons"""
77
- if not results:
78
- return "<p>No results to display</p>"
79
-
80
- html = """
81
- <div style="font-family: Arial, sans-serif;">
82
- <h3>Generated Alt Text Results</h3>
83
- <script>
84
- function copyToClipboard(text, buttonId) {
85
- navigator.clipboard.writeText(text).then(function() {
86
- const button = document.getElementById(buttonId);
87
- const originalText = button.innerHTML;
88
- button.innerHTML = '✓ Copied!';
89
- button.style.backgroundColor = '#4CAF50';
90
- setTimeout(function() {
91
- button.innerHTML = originalText;
92
- button.style.backgroundColor = '#008CBA';
93
- }, 2000);
94
- });
95
- }
96
- </script>
97
- <style>
98
- .result-container {
99
- border: 1px solid #ddd;
100
- border-radius: 8px;
101
- padding: 15px;
102
- margin: 10px 0;
103
- background-color: #f9f9f9;
104
- }
105
- .image-preview {
106
- max-width: 300px;
107
- max-height: 200px;
108
- border-radius: 5px;
109
- margin-bottom: 10px;
110
- }
111
- .alt-text-container {
112
- background-color: white;
113
- padding: 10px;
114
- border-radius: 5px;
115
- border: 1px solid #eee;
116
- margin: 10px 0;
117
- }
118
- .copy-button {
119
- background-color: #008CBA;
120
- color: white;
121
- border: none;
122
- padding: 8px 16px;
123
- border-radius: 4px;
124
- cursor: pointer;
125
- font-size: 14px;
126
- margin-top: 5px;
127
- }
128
- .copy-button:hover {
129
- background-color: #007B9A;
130
- }
131
  .filename {
132
  font-weight: bold;
133
  color: #333;
 
17
  # Process the image
18
  inputs = processor(image, return_tensors="pt")
19
 
20
+ # Generate caption with better parameters to avoid repetition
21
  with torch.no_grad():
22
+ out = model.generate(
23
+ **inputs,
24
+ max_length=50, # Reduced max length
25
+ min_length=5, # Minimum length
26
+ num_beams=3, # Reduced beam search
27
+ early_stopping=True,
28
+ no_repeat_ngram_size=2, # Prevent repetition
29
+ do_sample=True, # Add sampling
30
+ temperature=0.7, # Control randomness
31
+ top_p=0.9, # Nucleus sampling
32
+ repetition_penalty=1.2 # Penalize repetition
33
+ )
34
 
35
  # Decode the generated text
36
  caption = processor.decode(out[0], skip_special_tokens=True)
37
 
38
+ # Clean up repetitive text
39
+ caption = clean_repetitive_text(caption)
40
+
41
  # Clean up the caption to make it more suitable as alt text
42
  if caption.lower().startswith("a picture of"):
43
  caption = caption[12:].strip()
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
+ def clean_repetitive_text(text):
68
+ """Clean repetitive text patterns"""
69
+ import re
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:
 
131
 
132
  return results
133
 
134
+
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
135
  .filename {
136
  font-weight: bold;
137
  color: #333;