app.py

import gradio as gr
import time
import spaces
from PIL import Image
from transformers import Qwen3VLForConditionalGeneration, Qwen2VLForConditionalGeneration, AutoProcessor, AutoModelForImageTextToText
from qwen_vl_utils import process_vision_info
import torch
import uuid
import os
import numpy as np

# Model configurations
MODEL_CONFIGS = {

    "KATIB OCR 0.8B 0.1": {
        "name": "oddadmix/Katib-Qwen3.5-0.8B-0.3",
        "class": AutoModelForImageTextToText,
        "prompt": "Free OCR.",
        "use_qwen3": True
    },
    "Qari OCR 0.2.2.1": {
        "name": "oddadmix/Qari-OCR-0.2.2.1-VL-2B-Instruct-merged",
        "class": Qwen2VLForConditionalGeneration,
        "prompt": "Below is the image of one page of a document, as well as some raw textual content that was previously extracted for it. Just return the plain text representation of this document as if you were reading it naturally. Do not hallucinate.",
        "use_qwen3": False
    }
}

# Load models
models = {}
processors = {}

for model_key, config in MODEL_CONFIGS.items():
    print(f"Loading {model_key}...")
    models[model_key] = config["class"].from_pretrained(
        config["name"],
        torch_dtype="auto",
        device_map="cuda"
    )
    processors[model_key] = AutoProcessor.from_pretrained(config["name"])

max_tokens = 2000

def resizeImage(image):
    if image.height > 1500:
        image = image.resize((int(image.width * 1500 / image.height), 1500), Image.Resampling.LANCZOS)
    return image

@spaces.GPU
def perform_ocr(image, model_choice):
    inputArray = np.any(image)
    if inputArray == False:
        return "Error Processing"
    
    """Process image and extract text using selected OCR model"""
    image = Image.fromarray(image)
    
    # Get model configuration
    config = MODEL_CONFIGS[model_choice]
    model = models[model_choice]
    processor = processors[model_choice]
    prompt = config["prompt"]
    use_qwen3 = config["use_qwen3"]
    
    # Resize image for Qwen3 model

    # image = resizeImage(image)
    print("Image resized")
    
    src = str(uuid.uuid4()) + ".png"
    image.save(src)
    print(src)
    # Prepare messages based on model type
    if use_qwen3:
        messages = [
            {
                "role": "user",
                "content": [
                    {"type": "image", "image": f"./{src}"},
                    {"type": "text", "text": prompt},
                ],
            }
        ]
    else:
        messages = [
            {
                "role": "user",
                "content": [
                    {"type": "image", "image": f"file://{src}"},
                    {"type": "text", "text": prompt},
                ],
            }
        ]
    
    # Process inputs based on model type
    if use_qwen3:
        inputs = processor.apply_chat_template(
            messages,
            tokenize=True,
            add_generation_prompt=True,
            return_dict=True,
            return_tensors="pt"
        )
        inputs = inputs.to(model.device)
    else:
        text = processor.apply_chat_template(
            messages, tokenize=False, add_generation_prompt=True
        )
        image_inputs, video_inputs = process_vision_info(messages)
        inputs = processor(
            text=[text],
            images=image_inputs,
            videos=video_inputs,
            padding=True,
            return_tensors="pt",
        )
        inputs = inputs.to("cuda")
    
    # Generate text
    generated_ids = model.generate(**inputs, max_new_tokens=max_tokens, use_cache=True)
    generated_ids_trimmed = [
        out_ids[len(in_ids):] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
    ]
    output_text = processor.batch_decode(
        generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
    )[0]
    
    # Cleanup
    os.remove(src)
    return output_text

# Create Gradio interface
with gr.Blocks(title="Arabic OCR Models Demo") as demo:
    gr.Markdown("# Arabic OCR Models Demo")
    gr.Markdown("Upload an image to extract Arabic text in real-time. Choose between different OCR models.")
    
    with gr.Row():
        with gr.Column(scale=1):
            # Model selection dropdown
            model_dropdown = gr.Dropdown(
                choices=list(MODEL_CONFIGS.keys()),
                value=list(MODEL_CONFIGS.keys())[0],
                label="Select OCR Model",
                interactive=True
            )
            
            # Input image
            image_input = gr.Image(type="numpy", label="Upload Image")
            
            # Example gallery
            gr.Examples(
                examples=[
                    ["0.4.png"],
                    ["2.jpg"],
                    ["3.jpg"]
                ],
                inputs=image_input,
                label="Example Images",
                examples_per_page=4
            )
            
            # Submit button
            submit_btn = gr.Button("Extract Text")
        
        with gr.Column(scale=1):
            # Output text
            output = gr.Textbox(label="Extracted Text", lines=20, show_copy_button=True)
            
            # Model details
            with gr.Accordion("Model Information", open=False):
                gr.Markdown("""
                **Available Models:**
                
                1. **KATIB OCR 0.1 0.8B **
                   - Model: oddadmix/Katib-Qwen3.5-0.8B-0.1
                   - Based on Qwen3.5
                   - Size: 0.8B parameters
                
                2. **Qari OCR 0.2.2.1**
                   - Model: NAMAA-Space/Qari-OCR-0.2.2.1-VL-2B-Instruct
                   - Based on Qwen2-VL architecture
                   - Size: 2B parameters
                
                **Context window:** Supports up to 2000 output tokens
                """)
    
    # Set up processing flow
    submit_btn.click(fn=perform_ocr, inputs=[image_input, model_dropdown], outputs=output)
    image_input.change(fn=perform_ocr, inputs=[image_input, model_dropdown], outputs=output)

demo.launch()