Workpam commited on
Commit
ea537b3
·
verified ·
1 Parent(s): f6ac175

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +43 -48
app.py CHANGED
@@ -1,75 +1,70 @@
1
  import gradio as gr
2
- from transformers import pipeline
3
- from PIL import Image
4
  import torch
 
 
5
 
6
- # Device setup (GPU if available)
7
  device = "cuda" if torch.cuda.is_available() else "cpu"
8
 
9
- # Lightweight, reliable captioning pipeline
10
- captioner = pipeline(
11
- task="image-to-text",
12
- model="nlpconnect/vit-gpt2-image-captioning",
13
- device=0 if device == "cuda" else -1
14
- )
 
 
 
15
 
 
16
  def generate_alt_text(files):
17
- """
18
- Generate alt-text captions for each uploaded image file.
19
- """
20
  captions = []
21
  for f in files:
22
- path = f.name if hasattr(f, "name") else f
23
  try:
24
- img = Image.open(path).convert("RGB")
25
- result = captioner(img)
26
- text = result[0].get("generated_text") or result[0].get("caption")
 
 
 
 
 
 
 
 
 
 
27
  except Exception as e:
28
  text = f"<Error processing {path}: {e}>"
29
  captions.append(text.strip())
 
30
  return "\n".join(f"{i+1}. {cap}" for i, cap in enumerate(captions))
31
 
32
- # Build the Gradio interface
33
- demo = gr.Blocks(title="Reliable Alt‑Text Generator")
34
  with demo:
35
- gr.Markdown("# Reliable Alt Text Generator")
36
- gr.Markdown("Upload multiple images, preview them, and get accurate captions.")
37
 
38
- # 1) File uploader
39
- file_input = gr.File(
40
- label="Upload Images",
41
- file_count="multiple",
42
- file_types=["image"]
43
- )
44
 
45
- # 2) Preview gallery (4 columns) — NO .style() here
46
- preview = gr.Gallery(label="Image Preview", columns=4)
47
-
48
- # 3) Controls
49
  with gr.Row():
50
  generate_btn = gr.Button("Generate Alt Text")
51
  clear_btn = gr.Button("Clear All")
52
  copy_btn = gr.Button("Copy to Clipboard")
53
 
54
- # 4) Output
55
- output = gr.Textbox(
56
- label="Generated Alt Texts",
57
- lines=10,
58
- elem_id="output_box"
59
- )
60
 
61
- # Event bindings
62
- file_input.upload(lambda files: files, inputs=file_input, outputs=preview)
63
- generate_btn.click(fn=generate_alt_text, inputs=file_input, outputs=output)
64
- clear_btn.click(lambda: ([], [], ""), inputs=None, outputs=[file_input, preview, output])
65
- copy_btn.click(
66
- fn=None,
67
- inputs=None,
68
- outputs=None,
69
- js="() => { navigator.clipboard.writeText(document.getElementById('output_box').value); }"
70
- )
71
 
72
- # (Optional) enable queuing of up to 10 requests
73
  demo.queue(max_size=10)
74
-
75
  demo.launch(server_name="0.0.0.0", server_port=7860)
 
1
  import gradio as gr
2
+ from transformers import BlipProcessor, BlipForConditionalGeneration
 
3
  import torch
4
+ from PIL import Image
5
+ import easyocr
6
 
7
+ # Device setup
8
  device = "cuda" if torch.cuda.is_available() else "cpu"
9
 
10
+ # OCR reader (English)
11
+ reader = easyocr.Reader(['en'], gpu=(device=='cuda'))
12
+
13
+ # Load BLIP (large) model for high-quality image captioning
14
+ processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-large")
15
+ model = BlipForConditionalGeneration.from_pretrained(
16
+ "Salesforce/blip-image-captioning-large",
17
+ torch_dtype=torch.float16 if device=='cuda' else torch.float32
18
+ ).to(device)
19
 
20
+ # Generate alt-text including OCR-detected text
21
  def generate_alt_text(files):
 
 
 
22
  captions = []
23
  for f in files:
24
+ path = f.name if hasattr(f, 'name') else f
25
  try:
26
+ # Load image
27
+ image = Image.open(path).convert('RGB')
28
+ # Extract visible text via OCR
29
+ ocr_results = reader.readtext(path, detail=0)
30
+ visible_text = ' '.join(ocr_results).strip() or '<no text detected>'
31
+ # Prepare a prompt instructing model to include visible text
32
+ prompt = f"Describe this image in one sentence, and mention this text: '{visible_text}'."
33
+ # Tokenize inputs
34
+ inputs = processor(images=image, text=prompt, return_tensors="pt").to(device)
35
+ # Generate caption
36
+ with torch.no_grad():
37
+ out = model.generate(**inputs, max_new_tokens=50)
38
+ text = processor.decode(out[0], skip_special_tokens=True)
39
  except Exception as e:
40
  text = f"<Error processing {path}: {e}>"
41
  captions.append(text.strip())
42
+ # Numbered list
43
  return "\n".join(f"{i+1}. {cap}" for i, cap in enumerate(captions))
44
 
45
+ # Build Gradio UI
46
+ demo = gr.Blocks()
47
  with demo:
48
+ gr.Markdown("# Accurate Alt Text Generator with OCR + BLIP")
49
+ gr.Markdown("Upload multiple images, preview them, and get concise descriptions including embedded text.")
50
 
51
+ file_input = gr.File(label="Upload Images", file_count="multiple", file_types=["image"])
52
+ gallery = gr.Gallery(label="Image Preview", columns=4)
 
 
 
 
53
 
 
 
 
 
54
  with gr.Row():
55
  generate_btn = gr.Button("Generate Alt Text")
56
  clear_btn = gr.Button("Clear All")
57
  copy_btn = gr.Button("Copy to Clipboard")
58
 
59
+ output = gr.Textbox(label="Generated Alt Texts", lines=10, elem_id="output_box")
 
 
 
 
 
60
 
61
+ # Bind events
62
+ file_input.upload(lambda files: files, inputs=file_input, outputs=gallery)
63
+ generate_btn.click(generate_alt_text, inputs=file_input, outputs=output)
64
+ clear_btn.click(lambda: ([], [], ""), inputs=None, outputs=[file_input, gallery, output])
65
+ copy_btn.click(None, None, None, js="() => { navigator.clipboard.writeText(document.getElementById('output_box').value); }")
 
 
 
 
 
66
 
67
+ # Optional: queueing
68
  demo.queue(max_size=10)
69
+ # Launch
70
  demo.launch(server_name="0.0.0.0", server_port=7860)