first commit

app.py

@@ -0,0 +1,197 @@
+import atexit
+import functools
+from queue import Queue
+from threading import Event, Thread
+from huggingface_hub import snapshot_download
+import os
+
+from paddleocr import PaddleOCR
+import gradio as gr
+
+
+CONCURRENCY_LIMIT = 4
+
+
+class PaddleOCRModelManager(object):
+    def __init__(self, num_workers, model_factory):
+        super().__init__()
+        self._model_factory = model_factory
+        self._queue = Queue()
+        self._workers = []
+        self._model_initialized_event = Event()
+        for _ in range(num_workers):
+            worker = Thread(target=self._worker, daemon=False)
+            worker.start()
+            self._model_initialized_event.wait()
+            self._model_initialized_event.clear()
+            self._workers.append(worker)
+
+    def infer(self, *args, **kwargs):
+        result_queue = Queue(maxsize=1)
+        self._queue.put((args, kwargs, result_queue))
+        success, payload = result_queue.get()
+        if success:
+            return payload
+        else:
+            raise payload
+
+    def close(self):
+        for _ in self._workers:
+            self._queue.put(None)
+        for worker in self._workers:
+            worker.join()
+
+    def _worker(self):
+        model = self._model_factory()
+        self._model_initialized_event.set()
+        while True:
+            item = self._queue.get()
+            if item is None:
+                break
+            args, kwargs, result_queue = item
+            try:
+                result = model.predict(*args, **kwargs)
+                result_queue.put((True, result))
+            except Exception as e:
+                result_queue.put((False, e))
+            finally:
+                self._queue.task_done()
+
+
+def download_model():
+    """Download the fine-tuned Tachiwin model from Hugging Face"""
+    model_repo = "PaddlePaddle/PaddleOCR-VL"  # Update this!
+    model_dir = "./tachiwin_model"
+    
+    print(f"Downloading Tachiwin model from {model_repo}...")
+    
+    snapshot_download(
+        repo_id=model_repo,
+        local_dir=model_dir,
+        local_dir_use_symlinks=False
+    )
+    
+    print(f"Model downloaded successfully to {model_dir}")
+    return model_dir
+
+
+def create_model():
+    """Initialize PaddleOCR-VL with the fine-tuned Tachiwin model"""
+    model_dir = download_model()
+    
+    # Using PaddleOCR in doc_parser mode for VL model
+    return PaddleOCR(
+        vl_rec_model_name="PaddleOCR-VL-0.9B",
+        vl_rec_model_dir=model_dir,
+        use_gpu=False,
+        show_log=False
+    )
+
+
+# Initialize model manager with 2 workers
+print("Initializing Tachiwin Indigenous Languages OCR...")
+model_manager = PaddleOCRModelManager(2, create_model)
+print("Model ready!")
+
+
+def close_model_manager():
+    model_manager.close()
+
+
+atexit.register(close_model_manager)
+
+
+def inference(img):
+    """Process image with OCR and return extracted text in markdown format"""
+    if img is None:
+        return "Please upload an image."
+    
+    try:
+        result = model_manager.infer(img)[0]
+        
+        if not result:
+            return "No text detected in the image."
+        
+        # Extract text and format as markdown table
+        output_lines = ["# Extracted Text\n"]
+        output_lines.append("| Text | Confidence |")
+        output_lines.append("|------|-----------|")
+        
+        for line in result:
+            text = line[1][0]
+            confidence = f"{line[1][1]:.2%}"
+            output_lines.append(f"| {text} | {confidence} |")
+        
+        return "\n".join(output_lines)
+        
+    except Exception as e:
+        return f"Error during OCR processing: {str(e)}"
+
+
+title = '🌎 Tachiwin Indigenous Languages OCR'
+
+description = '''
+### PaddleOCR-VL Fine-tuned for the 68 Indigenous Languages of Mexico
+
+This model represents a **world first in tech access and linguistic rights**, specifically trained to recognize 
+the diverse character and glyph repertoire of Mexico's 68 indigenous languages.
+
+**How to use:** Simply upload an image containing text in any Mexican indigenous language, and the model will 
+detect and recognize the text.
+
+🔗 [PaddleOCR Documentation](https://github.com/PaddlePaddle/PaddleOCR)
+'''
+
+examples = [
+    ['example_nahuatl.jpg'],
+    ['example_maya.jpg'],
+    ['example_zapoteco.jpg'],
+]
+
+example_labels = """
+### Example Images:
+| Image | Language | Description |
+|-------|----------|-------------|
+| example_nahuatl.jpg | Náhuatl | Classical Nahuatl text with traditional glyphs |
+| example_maya.jpg | Maya (Yucatec) | Contemporary Maya writing with diacritics |
+| example_zapoteco.jpg | Zapoteco (Istmo) | Zapotec text from Oaxaca region |
+"""
+
+css = ".output_image, .input_image {height: 40rem !important; width: 100% !important;}"
+
+gr.Interface(
+    inference,
+    [
+        gr.Image(type='filepath', label='Input'),
+    ],
+    gr.Markdown(label='Output'),
+    title=title,
+    description=description,
+    examples=examples,
+    cache_examples=False,
+    css=css,
+    concurrency_limit=CONCURRENCY_LIMIT,
+    article=f"""
+    {example_labels}
+    
+    ### About Tachiwin
+    
+    **Tachiwin** (from Totonac - "Language") is dedicated to bridging 
+    the digital divide for indigenous languages of Mexico through AI technology.
+    
+    ### Supported Language Families
+    
+    **Uto-Aztecan:** Náhuatl, Yaqui, Mayo, Huichol, Tepehuán, Tarahumara  
+    **Mayan:** Maya, Tzeltal, Tzotzil, Chol, Tojolabal, Q'anjob'al, Mam  
+    **Oto-Manguean:** Zapoteco, Mixteco, Otomí, Mazateco, Chinanteco, Triqui  
+    **Totonac-Tepehua:** Totonaco, Tepehua  
+    **Mixe-Zoque:** Mixe, Zoque, Popoluca  
+    **Other:** Purépecha, Huave, Seri, Kickapoo, Kiliwa  
+    
+    ...covering all 68 officially recognized indigenous languages of Mexico.
+    
+    ---
+    
+    Made with ❤️ for linguistic diversity and indigenous rights
+    """
+).launch(debug=False)

requirements.txt

@@ -0,0 +1,9 @@
+paddlepaddle==3.0.0
+paddleocr[doc-parser]>=3.0.0
+gradio>=4.0.0
+huggingface-hub>=0.19.0
+Pillow>=10.0.0
+opencv-python-headless>=4.8.0
+numpy>=1.23.0
+safetensors>=0.4.0
+transformers>=4.30.0