app.py

import os

# Redirect cache to a writable path inside container
os.environ["XDG_CACHE_HOME"] = "/tmp/.cache"

import gradio as gr
from impresso_pipelines.ocrqa import OCRQAPipeline

pipeline = OCRQAPipeline()

LANGUAGES = ["en", "de", "fr"]

# Example OCR text (German text with typical OCR errors)
EXAMPLE_TEXT = """Vieles Seltsame geschieht auf Erden :
Nichts Seltsameres sieht der Mond
Als das Glück, das im Knopfloch wohnt.
Zaubrisch faßt es den ernsten Mann.
Ohne nach Weib u. Kinjd zu fragen
Reitet er aus, nach dem Glück zu jagen,
Nur nacb ihm war stets sein Vegehr.
Aber neben ihm 1reitet der Dämon her
Des Ehrgeizes mit finsterer Tücke,
Und so jagt er zuletzt auf die Brücke,
Die über dem Abgrund, d:m nächtlich schwarzen
Jählings abbricht."""

def process_ocr_qa(text, lang_choice):
    try:
        lang = None if lang_choice == "Auto-detect" else lang_choice
        result = pipeline(text, language=lang, diagnostics=True)

        # Format the output for better readability
        if isinstance(result, dict):
            output_lines = []
            
            # Language detection
            if 'language' in result:
                output_lines.append(f"🌍 Language: {result['language']}")
            
            # Quality score
            if 'score' in result:
                score = result['score']
                score_emoji = "🟢" if score >= 0.8 else "🟡" if score >= 0.5 else "🔴"
                output_lines.append(f"{score_emoji} Quality Score: {score:.1f}")
            
            # Diagnostics section
            if 'diagnostics' in result and result['diagnostics']:
                diagnostics = result['diagnostics']
                
                # Model information
                if 'model_id' in diagnostics:
                    output_lines.append(f"🤖 Model: {diagnostics['model_id']}")
                
                # Known tokens
                if 'known_tokens' in diagnostics and diagnostics['known_tokens']:
                    known_tokens = diagnostics['known_tokens']
                    output_lines.append(f"✅ Known tokens ({len(known_tokens)}): {', '.join(known_tokens)}")
                
                # Unknown tokens (potential OCR errors)
                if 'unknown_tokens' in diagnostics and diagnostics['unknown_tokens']:
                    unknown_tokens = diagnostics['unknown_tokens']
                    output_lines.append(f"❌ Potential OCR errors ({len(unknown_tokens)}): {', '.join(unknown_tokens)}")
                elif 'unknown_tokens' in diagnostics:
                    output_lines.append("✨ No potential OCR errors detected!")
            
            # Other fields
            for key, value in result.items():
                if key not in ['language', 'score', 'diagnostics']:
                    output_lines.append(f"🔍 {key.replace('_', ' ').title()}: {value}")
            
            return "\n\n".join(output_lines)
        else:
            return f"✨ Processed Result:\n{result}"
            
    except Exception as e:
        print("❌ Pipeline error:", e)
        return f"Error: {e}"

# Create the interface with logo and improved description
with gr.Blocks(title="OCR QA Demo") as demo:
    gr.HTML(
    """
    <a href="https://impresso-project.ch" target="_blank">
        <img src="https://huggingface.co/spaces/impresso-project/ocrqa-demo/resolve/main/logo.jpeg" alt="Impresso Project Logo" style="height: 100px;">
    </a>
    """
)
    gr.Markdown(
        """
    # 🔍 OCR Quality Assessment Demo

    This demo showcases the **OCR Quality Assessment (OCRQA)** pipeline developed as part of the [Impresso Project](https://impresso-project.ch). The pipeline evaluates the quality of text extracted via **Optical Character Recognition (OCR)** by estimating the proportion of recognizable words.

    It returns:
    - a **quality score** between **0.0 (poor)** and **1.0 (excellent)**, and
    - a list of **potential OCR errors** (unrecognized tokens).

    You can try the example below (a German text containing typical OCR errors), or paste your own OCR-processed text to assess its quality.
    """
    )
    
    with gr.Row():
        with gr.Column():
            text_input = gr.Textbox(
                label="Enter OCR Text", 
                value=EXAMPLE_TEXT,
                lines=8,
                placeholder="Enter your OCR text here..."
            )
            lang_dropdown = gr.Dropdown(
                choices=["Auto-detect"] + LANGUAGES, 
                value="de", 
                label="Language"
            )
            submit_btn = gr.Button("🔍 Analyze OCR Quality", variant="primary")
        
        with gr.Column():
            with gr.Row():
                output = gr.Textbox(
                    label="Analysis Results", 
                    lines=15,
                    placeholder="Results will appear here...",
                    scale=10
                )
                info_btn = gr.Button("Pipeline Info", size="sm", scale=1)
    
    # Info modal/accordion for pipeline details
    with gr.Accordion("📝 About the OCR QA Pipeline", open=False, visible=False) as info_accordion:
        gr.Markdown(
            """
            - **Quality Score**: Evaluates the overall quality of OCR text. From 0.0 (poor) to 1.0 (excellent)
            - **Known tokens**: Words recognized as valid in the selected language
            - **Potential OCR errors**: Identifies common OCR mistakes and artifacts
            """
        )
    
    submit_btn.click(
        fn=process_ocr_qa,
        inputs=[text_input, lang_dropdown],
        outputs=output
    )
    
    # Toggle info visibility when info button is clicked
    info_btn.click(
        fn=lambda: gr.Accordion(visible=True, open=True),
        outputs=info_accordion
    )

demo.launch(server_name="0.0.0.0", server_port=7860)