Upload 5 files

.gitattributes

@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+32600926.jpg filter=lfs diff=lfs merge=lfs -text
+test.jpg filter=lfs diff=lfs merge=lfs -text

app.py

@@ -0,0 +1,283 @@
+import gradio as gr
+import spaces
+from transformers import Qwen2VLForConditionalGeneration, AutoProcessor, Qwen2_5_VLForConditionalGeneration
+from qwen_vl_utils import process_vision_info
+import torch
+from PIL import Image
+import subprocess
+from datetime import datetime
+import numpy as np
+import os
+from gliner import GLiNER
+import json
+import tempfile
+import zipfile
+import base64
+import io
+
+# Initialize GLiNER model
+gliner_model = GLiNER.from_pretrained("knowledgator/modern-gliner-bi-large-v1.0")
+
+DEFAULT_NER_LABELS = "person, organization, location, date, event"
+
+# subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
+
+# models = {
+#     "Qwen/Qwen2-VL-7B-Instruct": AutoModelForCausalLM.from_pretrained("Qwen/Qwen2-VL-7B-Instruct", trust_remote_code=True, torch_dtype="auto", _attn_implementation="flash_attention_2").cuda().eval()
+
+# }
+
+class TextWithMetadata(list):
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args)
+        self.original_text = kwargs.get('original_text', '')
+        self.entities = kwargs.get('entities', [])
+
+def array_to_image_path(image_array):
+    # Convert numpy array to PIL Image
+    img = Image.fromarray(np.uint8(image_array))
+    img.thumbnail((1024, 1024))
+    
+    # Generate a unique filename using timestamp
+    timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
+    filename = f"image_{timestamp}.png"
+    
+    # Save the image
+    img.save(filename)
+    
+    # Get the full path of the saved image
+    full_path = os.path.abspath(filename)
+    
+    return full_path
+    
+models = {
+    "Qwen/Qwen2.5-VL-7B-Instruct": Qwen2_5_VLForConditionalGeneration.from_pretrained("Qwen/Qwen2.5-VL-7B-Instruct", trust_remote_code=True, torch_dtype="auto").cuda().eval()
+
+}
+
+processors = {
+    "Qwen/Qwen2.5-VL-7B-Instruct": AutoProcessor.from_pretrained("Qwen/Qwen2.5-VL-7B-Instruct", trust_remote_code=True)
+}
+
+DESCRIPTION = "This demo uses[Qwen2.5-VL-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct)"
+
+kwargs = {}
+kwargs['torch_dtype'] = torch.bfloat16
+
+user_prompt = '<|user|>\n'
+assistant_prompt = '<|assistant|>\n'
+prompt_suffix = "<|end|>\n"
+
+@spaces.GPU
+def run_example(image, model_id="Qwen/Qwen2.5-VL-7B-Instruct", run_ner=False, ner_labels=DEFAULT_NER_LABELS):
+    # First get the OCR text
+    text_input = "Convert the image to text."
+    
+    # Handle various image input formats
+    if image is None:
+        raise ValueError("Image path is None.")
+    
+    # Case 1: Image is a dictionary with base64 data (from API calls)
+    if isinstance(image, dict) and 'data' in image and isinstance(image['data'], str):
+        if image['data'].startswith('data:image'):
+            # Extract the base64 part after the comma
+            base64_data = image['data'].split(',', 1)[1]
+            # Convert base64 to bytes and then to PIL Image
+            image_bytes = base64.b64decode(base64_data)
+            pil_image = Image.open(io.BytesIO(image_bytes)).convert("RGB")
+            # Convert to numpy array for further processing
+            image = np.array(pil_image)
+    
+    # Convert numpy array to image path
+    image_path = array_to_image_path(image)
+    
+    model = models[model_id]
+    processor = processors[model_id]
+    
+    prompt = f"{user_prompt}<|image_1|>\n{text_input}{prompt_suffix}{assistant_prompt}"
+    image = Image.fromarray(image).convert("RGB")
+    messages = [
+    {
+            "role": "user",
+            "content": [
+                {
+                    "type": "image",
+                    "image": image_path,
+                },
+                {"type": "text", "text": text_input},
+            ],
+        }
+    ]
+    
+    # Preparation for inference
+    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")
+    
+    # Inference: Generation of the output
+    generated_ids = model.generate(**inputs, max_new_tokens=1024)
+    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
+    )
+    
+    ocr_text = output_text[0]
+    
+    # If NER is enabled, process the OCR text
+    if run_ner:
+        ner_results = gliner_model.predict_entities(
+            ocr_text,
+            ner_labels.split(","),
+            threshold=0.3
+        )
+        
+        # Create a list of tuples (text, label) for highlighting
+        highlighted_text = []
+        last_end = 0
+        
+        # Sort entities by start position
+        sorted_entities = sorted(ner_results, key=lambda x: x["start"])
+        
+        # Process each entity and add non-entity text segments
+        for entity in sorted_entities:
+            # Add non-entity text before the current entity
+            if last_end < entity["start"]:
+                highlighted_text.append((ocr_text[last_end:entity["start"]], None))
+            
+            # Add the entity text with its label
+            highlighted_text.append((
+                ocr_text[entity["start"]:entity["end"]],
+                entity["label"]
+            ))
+            last_end = entity["end"]
+        
+        # Add any remaining text after the last entity
+        if last_end < len(ocr_text):
+            highlighted_text.append((ocr_text[last_end:], None))
+        
+        # Create TextWithMetadata instance with the highlighted text and metadata
+        result = TextWithMetadata(highlighted_text, original_text=ocr_text, entities=ner_results)
+        return result, result  # Return twice: once for display, once for state
+    
+    # If NER is disabled, return the text without highlighting
+    result = TextWithMetadata([(ocr_text, None)], original_text=ocr_text, entities=[])
+    return result, result  # Return twice: once for display, once for state
+
+
+with gr.Blocks(css=css) as demo:
+    # Add state variables to store OCR results
+    ocr_state = gr.State()
+    
+    gr.Image("Caracal.jpg", interactive=False)
+    with gr.Tab(label="Image Input", elem_classes="tabs"):
+        with gr.Row():
+            with gr.Column(elem_classes="input-container"):
+                input_img = gr.Image(label="Input Picture", elem_classes="gr-image-input")
+                model_selector = gr.Dropdown(choices=list(models.keys()), label="Model", value="Qwen/Qwen2.5-VL-7B-Instruct", elem_classes="gr-dropdown")
+                
+                # Add NER controls
+                with gr.Row():
+                    ner_checkbox = gr.Checkbox(label="Run Named Entity Recognition", value=False)
+                    ner_labels = gr.Textbox(
+                        label="NER Labels (comma-separated)", 
+                        value=DEFAULT_NER_LABELS,
+                        visible=False
+                    )
+                
+                submit_btn = gr.Button(value="Submit", elem_classes="submit-btn")
+            with gr.Column(elem_classes="output-container"):
+                output_text = gr.HighlightedText(label="Output Text", elem_id="output")
+
+        # Show/hide NER labels based on checkbox
+        ner_checkbox.change(
+            lambda x: gr.update(visible=x),
+            inputs=[ner_checkbox],
+            outputs=[ner_labels]
+        )
+        
+        # Modify the submit button click handler to update state
+        submit_btn.click(
+            run_example,
+            inputs=[input_img, model_selector, ner_checkbox, ner_labels],
+            outputs=[output_text, ocr_state]  # Add ocr_state to outputs
+        )
+    with gr.Row():
+        filename = gr.Textbox(label="Save filename (without extension)", placeholder="Enter filename to save")
+        download_btn = gr.Button("Download Image & Text", elem_classes="submit-btn")
+        download_output = gr.File(label="Download")
+
+    # Modify create_zip to use the state data
+    def create_zip(image, fname, ocr_result):
+        # Validate inputs
+        if not fname or image is None:  # Changed the validation check
+            return None
+        
+        try:
+            # Convert numpy array to PIL Image if needed
+            if isinstance(image, np.ndarray):
+                image = Image.fromarray(image)
+            elif not isinstance(image, Image.Image):
+                return None
+            
+            with tempfile.TemporaryDirectory() as temp_dir:
+                # Save image
+                img_path = os.path.join(temp_dir, f"{fname}.png")
+                image.save(img_path)
+                
+                # Use the OCR result from state
+                original_text = ocr_result.original_text if ocr_result else ""
+                entities = ocr_result.entities if ocr_result else []
+                
+                # Save text
+                txt_path = os.path.join(temp_dir, f"{fname}.txt")
+                with open(txt_path, 'w', encoding='utf-8') as f:
+                    f.write(original_text)
+                
+                # Create JSON with text and entities
+                json_data = {
+                    "text": original_text,
+                    "entities": entities,
+                    "image_file": f"{fname}.png"
+                }
+                
+                # Save JSON
+                json_path = os.path.join(temp_dir, f"{fname}.json")
+                with open(json_path, 'w', encoding='utf-8') as f:
+                    json.dump(json_data, f, indent=2, ensure_ascii=False)
+                
+                # Create zip file
+                output_dir = "downloads"
+                os.makedirs(output_dir, exist_ok=True)
+                zip_path = os.path.join(output_dir, f"{fname}.zip")
+                
+                with zipfile.ZipFile(zip_path, 'w') as zipf:
+                    zipf.write(img_path, os.path.basename(img_path))
+                    zipf.write(txt_path, os.path.basename(txt_path))
+                    zipf.write(json_path, os.path.basename(json_path))
+                
+                return zip_path
+
+        except Exception as e:
+            print(f"Error creating zip: {str(e)}")
+            return None
+
+    # Update the download button click handler to include state
+    download_btn.click(
+        create_zip,
+        inputs=[input_img, filename, ocr_state],
+        outputs=[download_output]
+    )
+
+demo.queue(api_open=False)
+demo.launch(debug=True)

call.py

@@ -0,0 +1,73 @@
+import os
+import requests
+import json
+import dotenv
+from PIL import Image
+import io
+import base64
+import numpy as np
+from gradio_client import Client
+
+# Load environment variables from .env file
+dotenv.load_dotenv()
+
+# Get Hugging Face token from environment
+hf_token = os.getenv("HF_TOKEN")
+
+# Create client for the Hugging Face Space with authentication
+client = Client("wjbmattingly/caracal", hf_token=hf_token)
+
+# Example usage
+if __name__ == "__main__":
+    # Path to input image
+    image_path = "test.jpg"
+    
+    # Check if image exists
+    if not os.path.exists(image_path):
+        print(f"Error: Image file not found: {image_path}")
+        exit(1)
+    
+    # Print available API endpoints (for reference)
+    print("Available API endpoints:")
+    for endpoint in client.endpoints:
+        print(f"  - {endpoint}")
+    
+    print(f"\nProcessing image: {image_path}")
+    
+    # Convert the image to base64
+    with open(image_path, "rb") as image_file:
+        image_bytes = image_file.read()
+    
+    # Convert to base64 string
+    base64_image = base64.b64encode(image_bytes).decode("utf-8")
+    
+    # Format as expected by Gradio's ImageData model
+    image_data = {
+        "data": f"data:image/jpeg;base64,{base64_image}",
+        "is_file": False
+    }
+    
+    # Call the run_example function with the appropriate parameters
+    result = client.predict(
+        image_data,                         # Base64 encoded image
+        "Qwen/Qwen2.5-VL-7B-Instruct",      # Model selection
+        False,                              # NER disabled
+        "person, organization, location, date, event",  # Default NER labels
+        api_name="/run_example"             # API endpoint name
+    )
+    
+    # Process and display the result
+    if result:
+        print("\nExtracted Text:")
+        if isinstance(result, list):
+            full_text = ""
+            for segment in result:
+                if isinstance(segment, list) and len(segment) > 0:
+                    text = segment[0]
+                    full_text += text
+                    print(text, end="")
+            print("\n")
+        else:
+            print(result)
+    else:
+        print("No result returned from API")