app.py

import gradio as gr
from transformers import pipeline
from helper import render_results_in_image, summarize_predictions_natural_language, ignore_warnings

# Suppress non-critical warnings
ignore_warnings()

# Load DETR object detection pipeline with lower threshold
pipe = pipeline("object-detection", model="facebook/detr-resnet-50", threshold=0.3)

# Function to get predictions and process image + summary
def get_pipeline_prediction(pil_image):
    pipeline_output = pipe(pil_image)

    if not pipeline_output:
        return pil_image, "No objects detected."

    processed_image = render_results_in_image(pil_image, pipeline_output)
    summary = summarize_predictions_natural_language(pipeline_output)

    return processed_image, summary


# Gradio interface
iface = gr.Interface(
    fn=get_pipeline_prediction,
    inputs=gr.Image(label="Upload an Image", type="pil"),
    outputs=[
        gr.Image(label="Image with Detected Objects", type="pil"),
        gr.Textbox(label="Summary of Detected Objects")
    ],
    examples=[
        ["examples/kittens.jpeg"],
        ["examples/beach.jpeg"]  
    ],
    title="Object Detection with DETR",
    description="Upload an image to detect objects. Bounding boxes will be drawn and a natural language summary will be provided.",
    flagging_mode="never"
)

if __name__ == "__main__":
    iface.launch()