Workpam commited on
Commit
cb2eb58
·
verified ·
1 Parent(s): be0f145

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +81 -81
app.py CHANGED
@@ -7,101 +7,118 @@ from PIL import Image
7
  from concurrent.futures import ThreadPoolExecutor
8
  import pandas as pd
9
 
10
- # 1. Ensure the Replicate API token is set or crash early
11
- if "REPLICATE_API_TOKEN" not in os.environ:
12
- raise RuntimeError("Missing REPLICATE_API_TOKEN in environment!")
13
-
14
- # 2. Inject the token for all replicate calls
15
- replicate.api_token = os.environ["REPLICATE_API_TOKEN"]
16
-
17
- # 3. Initialize the Replicate client and preload BLIP model
18
- client = replicate.Client()
19
- blip_model = client.models.get("salesforce/blip-image-captioning-base")
20
-
21
- # 4. Initialize EasyOCR with GPU support if available
 
 
 
 
 
22
  gpu_flag = torch.cuda.is_available()
23
  ocr_reader = easyocr.Reader(["en"], gpu=gpu_flag)
24
 
25
- # 5. Load YOLOv5 model for object detection
26
- yolo_model = torch.hub.load("ultralytics/yolov5", "yolov5s", pretrained=True)
27
- if gpu_flag:
28
- yolo_model.to("cuda")
 
 
 
 
 
 
29
 
30
  def process_file(file_obj, max_len):
31
  """
32
- Process a single image file:
33
- - Open & verify image
34
- - Extract visible text via EasyOCR
35
- - Detect objects via YOLOv5
36
- - Generate base caption via BLIP
37
- - Combine & truncate to max_len
38
  """
39
- path = getattr(file_obj, 'name', file_obj)
40
- filename = os.path.basename(path)
 
 
 
41
 
42
- # Open & verify image
43
  try:
44
- with Image.open(path).convert('RGB') as img:
45
  img.verify()
46
  except Exception:
47
  return (filename, "<Error opening file>")
48
 
49
- # OCR extraction
50
  try:
51
- ocr_texts = ocr_reader.readtext(path, detail=0)
52
- visible_text = ' '.join(ocr_texts).strip()
53
  except Exception:
54
- visible_text = ''
55
-
56
- # Object detection
 
 
 
 
 
 
 
 
 
 
57
  try:
58
- results = yolo_model(path)
59
- labels = results.pandas().xyxy[0]['name'].unique().tolist()
60
- labels = ', '.join(labels[:5])
61
- except Exception:
62
- labels = ''
63
-
64
- # Generate base caption via BLIP
65
- try:
66
- with open(path, 'rb') as img_file:
67
- result = blip_model.predict(
68
- image=img_file,
69
  num_beams=5,
70
  max_length=75,
71
  min_length=5,
72
  no_repeat_ngram_size=2
73
  )
74
- base_caption = result[0] if isinstance(result, list) else str(result)
75
  except Exception as e:
76
  return (filename, f"Error generating caption: {e}")
77
 
78
- # Compose final caption
79
- parts = [base_caption.strip()]
80
  if labels:
81
  parts.append(f"Objects: {labels}")
82
- if visible_text:
83
- parts.append(f"Text: {visible_text}")
84
  caption = '; '.join(parts)
85
 
86
- # Truncate without cutting words
87
  if len(caption) > max_len:
88
  caption = caption[:max_len].rsplit(' ', 1)[0] + '…'
89
 
90
  return (filename, caption)
91
 
92
- # Determine available CPU cores for parallel processing
93
- cpu_cores = os.cpu_count() or 4
94
 
95
  def generate_alt_text(files, max_len):
96
  """
97
- Generate alt text in parallel and return a DataFrame & CSV.
98
  """
 
 
 
 
 
 
 
 
99
  if not files:
100
  return pd.DataFrame(columns=["filename", "alt_text"]), ""
101
 
102
  max_workers = min(len(files), cpu_cores)
103
- with ThreadPoolExecutor(max_workers=max_workers) as executor:
104
- results = list(executor.map(lambda f: process_file(f, max_len), files))
105
 
106
  df = pd.DataFrame(results, columns=["filename", "alt_text"])
107
  return df, df.to_csv(index=False)
@@ -109,54 +126,37 @@ def generate_alt_text(files, max_len):
109
 
110
  def download_csv(csv_str):
111
  """
112
- Save CSV string to a file and return its path for download.
113
  """
114
  path = "alt_texts.csv"
115
  with open(path, 'w', encoding='utf-8') as f:
116
  f.write(csv_str)
117
  return path
118
 
119
- # Build the Gradio interface
120
  demo = gr.Blocks(title="AI-Powered Alt Text Generator")
121
  with demo:
122
- gr.Markdown("# AI-Powered Alt Text Generator")
123
  gr.Markdown(
124
- "Generate concise, context-rich alt text using OCR, object detection, "
125
- "and BLIP, all under a customizable character limit."
126
  )
127
 
128
  with gr.Row():
129
- file_input = gr.File(
130
- label="Upload Images",
131
- file_count="multiple",
132
- file_types=["image"]
133
- )
134
- max_len = gr.Slider(
135
- minimum=50,
136
- maximum=200,
137
- step=5,
138
- value=125,
139
- label="Max Alt Text Length (chars)"
140
- )
141
 
142
- gallery = gr.Gallery(label="Image Preview", columns=4).style(grid=[4], height="auto")
143
 
144
  with gr.Row():
145
  gen_btn = gr.Button("Generate Alt Text")
146
  download_btn = gr.Button("Download CSV")
147
 
148
- output_df = gr.Dataframe(headers=["filename", "alt_text"], label="Results Table")
149
- download_file = gr.File(label="Download CSV File")
150
 
151
- # Event bindings
152
  file_input.upload(lambda files: files, inputs=file_input, outputs=gallery)
153
- gen_btn.click(
154
- fn=generate_alt_text,
155
- inputs=[file_input, max_len],
156
- outputs=[output_df, download_file]
157
- )
158
  download_btn.click(fn=download_csv, inputs=download_file, outputs=download_file)
159
 
160
- # Launch the app
161
  if __name__ == "__main__":
162
  demo.launch(server_name="0.0.0.0", server_port=7860, share=True)
 
7
  from concurrent.futures import ThreadPoolExecutor
8
  import pandas as pd
9
 
10
+ # Local development: load variables from .env if present
11
+ try:
12
+ from dotenv import load_dotenv
13
+ load_dotenv()
14
+ except ImportError:
15
+ pass # python-dotenv not installed
16
+
17
+ # On Hugging Face Spaces, set this in Settings → Secrets and Variables
18
+ REPLICATE_TOKEN = os.getenv("REPLICATE_API_TOKEN", "")
19
+ replicate.api_token = REPLICATE_TOKEN
20
+
21
+ # Initialize models only if token is present
22
+ token_available = bool(REPLICATE_TOKEN)
23
+ client = replicate.Client() if token_available else None
24
+ blip_model = client.models.get("salesforce/blip-image-captioning-base") if token_available else None
25
+
26
+ # OCR setup
27
  gpu_flag = torch.cuda.is_available()
28
  ocr_reader = easyocr.Reader(["en"], gpu=gpu_flag)
29
 
30
+ # (Optional) Object detection using YOLOv5
31
+ use_yolo = True
32
+ if use_yolo:
33
+ yolo_model = torch.hub.load("ultralytics/yolov5", "yolov5s", pretrained=True)
34
+ if gpu_flag:
35
+ yolo_model.to("cuda")
36
+
37
+ # CPU cores for parallelism
38
+ cpu_cores = os.cpu_count() or 4
39
+
40
 
41
  def process_file(file_obj, max_len):
42
  """
43
+ Process a single image: OCR, object detection, BLIP captioning, and truncation.
 
 
 
 
 
44
  """
45
+ filename = os.path.basename(getattr(file_obj, 'name', file_obj))
46
+
47
+ # If no token, skip processing
48
+ if not token_available:
49
+ return (filename, "Error: REPLICATE_API_TOKEN not set. Add it in Settings → Secrets and Variables.")
50
 
51
+ # Verify image
52
  try:
53
+ with Image.open(getattr(file_obj, 'name', file_obj)).convert('RGB') as img:
54
  img.verify()
55
  except Exception:
56
  return (filename, "<Error opening file>")
57
 
58
+ # Extract visible text via OCR
59
  try:
60
+ texts = ocr_reader.readtext(getattr(file_obj, 'name', file_obj), detail=0)
61
+ visible = ' '.join(texts).strip()
62
  except Exception:
63
+ visible = ''
64
+
65
+ # Detect objects via YOLOv5
66
+ labels = ''
67
+ if use_yolo:
68
+ try:
69
+ res = yolo_model(getattr(file_obj, 'name', file_obj))
70
+ labels_list = res.pandas().xyxy[0]['name'].unique().tolist()[:5]
71
+ labels = ', '.join(labels_list)
72
+ except Exception:
73
+ labels = ''
74
+
75
+ # Generate caption via BLIP
76
  try:
77
+ with open(getattr(file_obj, 'name', file_obj), 'rb') as img_f:
78
+ output = blip_model.predict(
79
+ image=img_f,
 
 
 
 
 
 
 
 
80
  num_beams=5,
81
  max_length=75,
82
  min_length=5,
83
  no_repeat_ngram_size=2
84
  )
85
+ base = output[0] if isinstance(output, list) else str(output)
86
  except Exception as e:
87
  return (filename, f"Error generating caption: {e}")
88
 
89
+ # Assemble parts
90
+ parts = [base.strip()]
91
  if labels:
92
  parts.append(f"Objects: {labels}")
93
+ if visible:
94
+ parts.append(f"Text: {visible}")
95
  caption = '; '.join(parts)
96
 
97
+ # Truncate cleanly
98
  if len(caption) > max_len:
99
  caption = caption[:max_len].rsplit(' ', 1)[0] + '…'
100
 
101
  return (filename, caption)
102
 
 
 
103
 
104
  def generate_alt_text(files, max_len):
105
  """
106
+ Run processing in parallel and return a DataFrame + CSV for download.
107
  """
108
+ # Handle no-huggingface-token case
109
+ if not token_available:
110
+ df = pd.DataFrame(
111
+ [("", "Error: REPLICATE_API_TOKEN not set. Add it in Settings → Secrets and Variables.")],
112
+ columns=["filename", "alt_text"]
113
+ )
114
+ return df, df.to_csv(index=False)
115
+
116
  if not files:
117
  return pd.DataFrame(columns=["filename", "alt_text"]), ""
118
 
119
  max_workers = min(len(files), cpu_cores)
120
+ with ThreadPoolExecutor(max_workers=max_workers) as exe:
121
+ results = list(exe.map(lambda f: process_file(f, max_len), files))
122
 
123
  df = pd.DataFrame(results, columns=["filename", "alt_text"])
124
  return df, df.to_csv(index=False)
 
126
 
127
  def download_csv(csv_str):
128
  """
129
+ Save and return path for CSV download.
130
  """
131
  path = "alt_texts.csv"
132
  with open(path, 'w', encoding='utf-8') as f:
133
  f.write(csv_str)
134
  return path
135
 
136
+ # Build Gradio app
137
  demo = gr.Blocks(title="AI-Powered Alt Text Generator")
138
  with demo:
139
+ gr.Markdown("# AI-Powered Alt Text Generator (Hugging Face Spaces)")
140
  gr.Markdown(
141
+ "Generate rich alt text with OCR, object detection, and BLIP—all under your custom length."
 
142
  )
143
 
144
  with gr.Row():
145
+ file_input = gr.File(label="Upload Images", file_count="multiple", file_types=["image"])
146
+ max_len = gr.Slider(50, 200, step=5, value=125, label="Max Alt Text Length (chars)")
 
 
 
 
 
 
 
 
 
 
147
 
148
+ gallery = gr.Gallery(label="Preview", columns=4).style(grid=[4], height="auto")
149
 
150
  with gr.Row():
151
  gen_btn = gr.Button("Generate Alt Text")
152
  download_btn = gr.Button("Download CSV")
153
 
154
+ output_df = gr.Dataframe(headers=["filename", "alt_text"], label="Results")
155
+ download_file = gr.File(label="CSV File")
156
 
 
157
  file_input.upload(lambda files: files, inputs=file_input, outputs=gallery)
158
+ gen_btn.click(fn=generate_alt_text, inputs=[file_input, max_len], outputs=[output_df, download_file])
 
 
 
 
159
  download_btn.click(fn=download_csv, inputs=download_file, outputs=download_file)
160
 
 
161
  if __name__ == "__main__":
162
  demo.launch(server_name="0.0.0.0", server_port=7860, share=True)