app.py

import gradio as gr
from transformers import AutoImageProcessor, AutoModelForObjectDetection
import torch

# Load the processor and model for table structure recognition
processor = AutoImageProcessor.from_pretrained("microsoft/table-transformer-structure-recognition")
model = AutoModelForObjectDetection.from_pretrained("microsoft/table-transformer-structure-recognition")

# Define the inference function
def predict(image):
    # Preprocess the input image
    inputs = processor(images=image, return_tensors="pt")
    
    # Perform object detection using the model
    with torch.no_grad():
        outputs = model(**inputs)

    # Extract bounding boxes and class labels
    predicted_boxes = outputs.pred_boxes[0].cpu().numpy()  # First image
    predicted_class_logits = outputs.logits[0].cpu().numpy()  # Class logits for the first image
    predicted_classes = predicted_class_logits.argmax(-1)  # Get class predictions
    class_names = model.config.id2label  # Get the class name mapping
    
    # Filter predictions to only include columns based on class name
    column_boxes = []
    for idx, class_id in enumerate(predicted_classes):
        class_name = class_names[class_id]
        if "column" in class_name.lower():  # Check if the class name contains 'column'
            column_boxes.append(predicted_boxes[idx])

    # Return the bounding boxes for columns
    return {"boxes": column_boxes, "classes": ["column"] * len(column_boxes)}

# Set up the Gradio interface
interface = gr.Interface(
    fn=predict,  # The function that gets called when an image is uploaded
    inputs=gr.Image(type="pil"),  # Image input (as PIL image)
    outputs="json",  # Outputting a JSON with the boxes and classes
)

# Launch the Gradio app
interface.launch()