File size: 6,471 Bytes
70bc954
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
39c32e3
 
 
 
 
 
70bc954
39c32e3
 
70bc954
39c32e3
e052503
39c32e3
 
70bc954
39c32e3
 
 
 
 
 
 
 
 
70bc954
39c32e3
 
70bc954
39c32e3
70bc954
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
39c32e3
 
 
 
70bc954
39c32e3
70bc954
 
 
 
 
 
 
 
39c32e3
70bc954
 
 
 
 
 
 
 
 
 
 
 
39c32e3
70bc954
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
39c32e3
 
 
 
70bc954
 
 
39c32e3
 
70bc954
 
39c32e3
70bc954
 
39c32e3
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
# import gradio as gr
# from transformers import AutoModel, AutoTokenizer
# import torch
# import tempfile
# import os
# import time

# # ------------------------------------------------------
# # 1. Load the CPU-Patched Model
# # ------------------------------------------------------
# # This is the specific repo that fixes the "Found no NVIDIA driver" error.
# MODEL_ID = "srimanth-d/GOT_CPU"

# print(f"⏳ Loading {MODEL_ID}...")

# # Load Tokenizer
# tokenizer = AutoTokenizer.from_pretrained(
#     MODEL_ID, 
#     trust_remote_code=True
# )

# # Load Model
# # low_cpu_mem_usage=True is safe here because this repo is patched for CPU.
# model = AutoModel.from_pretrained(
#     MODEL_ID, 
#     trust_remote_code=True, 
#     low_cpu_mem_usage=True, 
#     device_map='cpu', 
#     use_safetensors=True, 
#     pad_token_id=tokenizer.eos_token_id
# )

# model = model.eval().float() 
# print(f"✅ {MODEL_ID} Loaded! Ready for handwriting.")

# # ------------------------------------------------------
# # 2. The OCR Logic
# # ------------------------------------------------------
# def run_fast_handwriting_ocr(input_image):
#     if input_image is None:
#         return "No image provided."

#     start_time = time.time()
    
#     # Save temp file (Model expects a file path)
#     with tempfile.NamedTemporaryFile(delete=False, suffix=".jpg") as tmp:
#         input_image.save(tmp.name)
#         img_path = tmp.name

#     try:
#         # OCR_TYPE='ocr' tells the model to just read text (no formatting/latex)
#         # This is the fastest mode.
#         res = model.chat(tokenizer, img_path, ocr_type='ocr')
        
#         elapsed = time.time() - start_time
#         return f"{res}\n\n--- ⏱️ Time taken: {elapsed:.2f}s ---"

#     except Exception as e:
#         return f"Error: {e}"
    
#     finally:
#         # Cleanup
#         if os.path.exists(img_path):
#             os.remove(img_path)

# # ------------------------------------------------------
# # 3. Gradio Interface
# # ------------------------------------------------------
# with gr.Blocks(title="Fast Handwriting OCR") as demo:
#     gr.Markdown(f"## ✍️ Fast Handwriting OCR (GOT-OCR2.0)")
#     gr.Markdown("A specialized ~600M param model designed to read messy text quickly on CPU.")
    
#     with gr.Row():
#         input_img = gr.Image(type="pil", label="Upload Handwritten Note")
        
#     with gr.Row():
#         btn = gr.Button("Read Handwriting", variant="primary")
        
#     with gr.Row():
#         out_text = gr.Textbox(label="Recognized Text", lines=15)

#     btn.click(fn=run_fast_handwriting_ocr, inputs=input_img, outputs=out_text)

# if __name__ == "__main__":
#     demo.launch()




import gradio as gr
from transformers import AutoModel, AutoTokenizer
import torch
import tempfile
import os
import time
from PIL import Image

# ------------------------------------------------------
# 1. Load the Model (CPU Optimized)
# ------------------------------------------------------
MODEL_ID = "srimanth-d/GOT_CPU"

print(f"⏳ Loading {MODEL_ID}...")
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True)
model = AutoModel.from_pretrained(
    MODEL_ID, 
    trust_remote_code=True, 
    low_cpu_mem_usage=True, 
    device_map='cpu', 
    use_safetensors=True, 
    pad_token_id=tokenizer.eos_token_id
)
model = model.eval().float() 
print(f"✅ Model Loaded!")

# ------------------------------------------------------
# 2. Slicing Logic (The Fix)
# ------------------------------------------------------
def process_slice(img_slice, slice_index):
    """Save slice to temp file and run OCR"""
    with tempfile.NamedTemporaryFile(delete=False, suffix=f"_{slice_index}.jpg") as tmp:
        img_slice.save(tmp.name)
        slice_path = tmp.name
        
    try:
        # OCR_TYPE='ocr' is the fastest mode
        res = model.chat(tokenizer, slice_path, ocr_type='ocr')
        return res
    except Exception as e:
        return f"[Error in slice {slice_index}: {e}]"
    finally:
        if os.path.exists(slice_path):
            os.remove(slice_path)

def run_sliced_ocr(input_image):
    if input_image is None:
        return "No image provided."

    start_time = time.time()
    w, h = input_image.size
    
    # Heuristic: If image is tall, split it.
    # 1024 is the model's native resolution.
    full_text = ""
    
    # A. Smart Slicing Strategy
    # If the image is a standard document (Height > Width), slice vertically.
    if h > 1024:
        print(f"--- Slicing Image ({w}x{h}) ---")
        
        # Define 3 overlapping slices to cover a full A4 page nicely
        # Top half, Middle (to catch text on the fold), Bottom half
        slices = []
        
        # Slice 1: Top 40%
        slices.append(input_image.crop((0, 0, w, int(h * 0.40))))
        
        # Slice 2: Middle 40% (overlapping top and bottom)
        slices.append(input_image.crop((0, int(h * 0.30), w, int(h * 0.70))))
        
        # Slice 3: Bottom 40%
        slices.append(input_image.crop((0, int(h * 0.60), w, h)))

        results = []
        for i, sl in enumerate(slices):
            print(f"Processing slice {i+1}/3...")
            txt = process_slice(sl, i)
            results.append(txt)
            
        # Join with separators
        full_text = "\n--- [Top Section] ---\n" + results[0] + \
                    "\n--- [Middle Section] ---\n" + results[1] + \
                    "\n--- [Bottom Section] ---\n" + results[2]
                    
    else:
        # B. Small Image? Just run once.
        print("--- Processing Full Image ---")
        full_text = process_slice(input_image, 0)

    elapsed = time.time() - start_time
    return f"{full_text}\n\n--- ⏱️ Total Time: {elapsed:.2f}s ---"

# ------------------------------------------------------
# 3. Gradio Interface
# ------------------------------------------------------
with gr.Blocks(title="High-Res Handwriting OCR") as demo:
    gr.Markdown("## ✍️ Sliced Handwriting OCR")
    gr.Markdown("Splits the image into 3 chunks to maintain resolution for messy handwriting.")
    
    with gr.Row():
        input_img = gr.Image(type="pil", label="Upload Document")
        out_text = gr.Textbox(label="Extracted Text", lines=20)
        
    btn = gr.Button("Run Sliced OCR", variant="primary")
    btn.click(fn=run_sliced_ocr, inputs=input_img, outputs=out_text)

if __name__ == "__main__":
    demo.launch()