Workpam commited on
Commit
e0e127c
·
verified ·
1 Parent(s): 16dba33

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +68 -41
app.py CHANGED
@@ -2,7 +2,9 @@ import os
2
  import gradio as gr
3
  import replicate
4
  import easyocr
 
5
  from PIL import Image
 
6
 
7
  # 1. Ensure the API token is set or crash early
8
  if "REPLICATE_API_TOKEN" not in os.environ:
@@ -11,59 +13,84 @@ if "REPLICATE_API_TOKEN" not in os.environ:
11
  # 2. Inject it into the replicate module so every call is authenticated
12
  replicate.api_token = os.environ["REPLICATE_API_TOKEN"]
13
 
14
- # 3. Initialize the Replicate client
15
  client = replicate.Client()
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
16
 
17
- # Initialize EasyOCR for visible text extraction
18
- detector_langs = ['en'] # add more languages if needed
19
- reader = easyocr.Reader(detector_langs, gpu=False)
20
 
21
  def generate_alt_text(files):
22
  """
23
- Generates alt text using Replicate's BLIP model and includes any visible text via OCR.
24
  """
25
- captions = []
26
- for f in files:
27
- # Attempt to open image
28
- try:
29
- path = f.name if hasattr(f, 'name') else f
30
- image = Image.open(path).convert('RGB')
31
- except Exception:
32
- captions.append(f"<Error opening file {f}>")
33
- continue
34
-
35
- # OCR extraction
36
- try:
37
- ocr_list = reader.readtext(path, detail=0)
38
- visible_text = ' '.join(ocr_list).strip() or '<no text>'
39
- except Exception:
40
- visible_text = '<ocr error>'
41
-
42
- # Generate base caption via Replicate BLIP
43
- try:
44
- model = client.models.get("salesforce/blip-image-captioning-base")
45
- result = model.predict(image=open(path, "rb"))
46
- base_caption = result[0] if isinstance(result, list) else str(result)
47
- except Exception as e:
48
- captions.append(f"Error generating caption: {e}")
49
- continue
50
-
51
- # Compose final caption
52
- final = f"{base_caption.strip()} (Text: {visible_text})"
53
- captions.append(final)
54
-
55
- # Format numbered list
56
- return "\n".join([f"{i+1}. {c}" for i, c in enumerate(captions)])
57
 
58
  # Build Gradio interface
59
  demo = gr.Blocks(title="Replicate‑Powered Alt Text Generator")
60
  with demo:
61
  gr.Markdown("# Alt Text Generator with Replicate BLIP")
62
  gr.Markdown(
63
- "Upload multiple images, and get detailed alt text including visible in‑image text. Powered by Replicate and EasyOCR."
 
64
  )
65
 
66
- # File upload
67
  file_input = gr.File(
68
  label="Upload Images",
69
  file_count="multiple",
@@ -79,10 +106,10 @@ with demo:
79
  clr_btn = gr.Button("Clear All")
80
  copy_btn = gr.Button("Copy to Clipboard")
81
 
82
- # Output
83
  output = gr.Textbox(label="Alt Texts", lines=10, elem_id="output_box")
84
 
85
- # Event bindings
86
  file_input.upload(lambda files: files, inputs=file_input, outputs=gallery)
87
  gen_btn.click(fn=generate_alt_text, inputs=file_input, outputs=output)
88
  clr_btn.click(lambda: ([], [], ""), inputs=None, outputs=[file_input, gallery, output])
 
2
  import gradio as gr
3
  import replicate
4
  import easyocr
5
+ import torch
6
  from PIL import Image
7
+ from concurrent.futures import ThreadPoolExecutor
8
 
9
  # 1. Ensure the API token is set or crash early
10
  if "REPLICATE_API_TOKEN" not in os.environ:
 
13
  # 2. Inject it into the replicate module so every call is authenticated
14
  replicate.api_token = os.environ["REPLICATE_API_TOKEN"]
15
 
16
+ # 3. Initialize the Replicate client and preload the BLIP model
17
  client = replicate.Client()
18
+ model = client.models.get("salesforce/blip-image-captioning-base")
19
+
20
+ # 4. Initialize EasyOCR with GPU support if available
21
+ gpu_flag = torch.cuda.is_available()
22
+ reader = easyocr.Reader(["en"], gpu=gpu_flag)
23
+
24
+ # Determine available CPU cores for parallel processing
25
+ cpu_cores = os.cpu_count() or 4
26
+
27
+
28
+ def process_file(file_obj):
29
+ """
30
+ Process a single file: open image, extract visible text via OCR,
31
+ generate caption via Replicate BLIP, and compose final output.
32
+ """
33
+ # Resolve file path
34
+ path = getattr(file_obj, 'name', file_obj)
35
+
36
+ # Open and validate image
37
+ try:
38
+ with Image.open(path).convert('RGB') as img:
39
+ img.verify() # verify it's an image
40
+ except Exception:
41
+ return f"<Error opening file {path}>"
42
+
43
+ # OCR extraction
44
+ try:
45
+ ocr_texts = reader.readtext(path, detail=0)
46
+ visible_text = ' '.join(ocr_texts).strip() or '<no text>'
47
+ except Exception:
48
+ visible_text = '<ocr error>'
49
+
50
+ # Generate base caption via Replicate BLIP with beam search for accuracy
51
+ try:
52
+ with open(path, 'rb') as img_file:
53
+ result = model.predict(
54
+ image=img_file,
55
+ num_beams=5,
56
+ max_length=75,
57
+ min_length=5,
58
+ no_repeat_ngram_size=2
59
+ )
60
+ base_caption = result[0] if isinstance(result, list) else str(result)
61
+ except Exception as e:
62
+ return f"Error generating caption: {e}"
63
+
64
+ # Compose and return final caption
65
+ return f"{base_caption.strip()} (Text: {visible_text})"
66
 
 
 
 
67
 
68
  def generate_alt_text(files):
69
  """
70
+ Generates alt text for multiple files in parallel, improving throughput.
71
  """
72
+ if not files:
73
+ return "No files uploaded."
74
+
75
+ # Limit workers to number of files or CPU cores
76
+ max_workers = min(len(files), cpu_cores)
77
+ with ThreadPoolExecutor(max_workers=max_workers) as executor:
78
+ results = list(executor.map(process_file, files))
79
+
80
+ # Format as numbered list
81
+ return "\n".join(f"{i+1}. {caption}" for i, caption in enumerate(results))
82
+
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
83
 
84
  # Build Gradio interface
85
  demo = gr.Blocks(title="Replicate‑Powered Alt Text Generator")
86
  with demo:
87
  gr.Markdown("# Alt Text Generator with Replicate BLIP")
88
  gr.Markdown(
89
+ "Upload multiple images, and get detailed alt text including visible in‑image text."
90
+ " Powered by Replicate BLIP and EasyOCR, with parallel processing for speed."
91
  )
92
 
93
+ # File upload component
94
  file_input = gr.File(
95
  label="Upload Images",
96
  file_count="multiple",
 
106
  clr_btn = gr.Button("Clear All")
107
  copy_btn = gr.Button("Copy to Clipboard")
108
 
109
+ # Output textbox
110
  output = gr.Textbox(label="Alt Texts", lines=10, elem_id="output_box")
111
 
112
+ # Bind events
113
  file_input.upload(lambda files: files, inputs=file_input, outputs=gallery)
114
  gen_btn.click(fn=generate_alt_text, inputs=file_input, outputs=output)
115
  clr_btn.click(lambda: ([], [], ""), inputs=None, outputs=[file_input, gallery, output])