app.py

import os
import logging
import sys

# Function to get logging level from environment variable
def get_logging_level(default_level=logging.DEBUG):  # Default to DEBUG for detailed logs
    log_level_str = os.getenv('VISION_AGENT_LOG_LEVEL', '').upper()
    if log_level_str == 'DEBUG':
        return logging.DEBUG
    elif log_level_str == 'INFO':
        return logging.INFO
    elif log_level_str == 'WARNING':
        return logging.WARNING
    elif log_level_str == 'ERROR':
        return logging.ERROR
    elif log_level_str == 'CRITICAL':
        return logging.CRITICAL
    else:
        return default_level

# Set up logging before importing the library
logging_level = get_logging_level()
logging.basicConfig(stream=sys.stdout, level=logging_level)
_LOGGER = logging.getLogger(__name__)

# Explicitly set logging level for the vision-agent library
vision_agent_logger = logging.getLogger('vision_agent')
vision_agent_logger.setLevel(logging_level)

# Set logging level for Hugging Face libraries
hf_hub_logger = logging.getLogger('huggingface_hub')
hf_hub_logger.setLevel(logging_level)

datasets_logger = logging.getLogger('datasets')
datasets_logger.setLevel(logging_level)

# Print the logging level to verify it's set correctly
print(f"Logging level set to: {logging.getLevelName(logging_level)}")

from huggingface_hub import login
import time
import gradio as gr
from typing import *
from pillow_heif import register_heif_opener
register_heif_opener()
import vision_agent as va
from vision_agent.tools import register_tool, load_image, owl_v2, overlay_bounding_boxes, save_image

# Perform login using the token
hf_token = os.getenv("HF_TOKEN")
login(token=hf_token, add_to_git_credential=True)

import numpy as np
from PIL import Image

def detect_brain_tumor(image, seg_input, debug: bool = True):
    """
    Detects a brain tumor in the given image and returns the annotated image.

    Parameters:
        image: The input image (as numpy array provided by Gradio).
        seg_input: The segmentation input (not used in this function, but required for Gradio).
        debug (bool): Flag to enable logging for debugging purposes.

    Returns:
        tuple: (numpy array of image, list of (label, (x1, y1, x2, y2)) tuples)
    """
    if debug:
        _LOGGER.debug(f"Image received, shape: {image.shape}")

    # Step 2: Detect brain tumor using owl_v2
    prompt = "detect brain tumor"
    detections = owl_v2(prompt, image)
    if debug:
        _LOGGER.debug(f"Raw detections: {detections}")

    # Step 3: Overlay bounding boxes on the image
    image_with_bboxes = overlay_bounding_boxes(image, detections)
    if debug:
        _LOGGER.debug("Bounding boxes overlaid on the image")

    # Prepare annotations for AnnotatedImage output
    annotations = []
    for detection in detections:
        label = detection['label']
        score = detection['score']
        bbox = detection['bbox']
        x1, y1, x2, y2 = bbox
        # Convert normalized coordinates to pixel coordinates
        height, width = image.shape[:2]
        x1, y1, x2, y2 = int(x1*width), int(y1*height), int(x2*width), int(y2*height)
        annotations.append(((x1, y1, x2, y2), f"{label} {score:.2f}"))

    if debug:
        _LOGGER.debug(f"Annotations: {annotations}")

    # Convert image to numpy array if it's not already
    if isinstance(image_with_bboxes, Image.Image):
        image_with_bboxes = np.array(image_with_bboxes)

    return (image_with_bboxes, annotations)

INTRO_TEXT="# 🔬🧠 OmniScience -- Agentic Imaging Analysis 🤖🧫"

with gr.Blocks(css="style.css") as demo:
    gr.Markdown(INTRO_TEXT)
    with gr.Tab("Segment/Detect"):
        with gr.Row():
            with gr.Column():
                image = gr.Image(type="numpy")
                seg_input = gr.Text(label="Entities to Segment/Detect", value="detect brain tumor")
        
            with gr.Column():
                annotated_image = gr.AnnotatedImage(label="Output")

        seg_btn = gr.Button("Submit")    
        examples = [["./examples/194_jpg.rf.3e3dd592d034bb5ee27a978553819f42.jpg", "detect brain tumor"],
                    ["./examples/239_jpg.rf.3dcc0799277fb78a2ab21db7761ccaeb.jpg", "detect brain tumor"],
                    ["./examples/1385_jpg.rf.3c67cb92e2922dba0e6dba86f69df40b.jpg", "detect brain tumor"],
                    ["./examples/1491_jpg.rf.3c658e83538de0fa5a3f4e13d7d85f12.jpg", "detect brain tumor"],
                    ["./examples/1550_jpg.rf.3d067be9580ec32dbee5a89c675d8459.jpg", "detect brain tumor"],
                    ["./examples/2256_jpg.rf.3afd7903eaf3f3c5aa8da4bbb928bc19.jpg", "detect brain tumor"],
                    ["./examples/2871_jpg.rf.3b6eadfbb369abc2b3bcb52b406b74f2.jpg", "detect brain tumor"],
                    ["./examples/2921_jpg.rf.3b952f91f27a6248091e7601c22323ad.jpg", "detect brain tumor"],
                    ]
        gr.Examples(
            examples=examples,
            inputs=[image, seg_input],
        )
        seg_inputs = [
            image,
            seg_input
            ]
        seg_outputs = [
            annotated_image
        ]
        seg_btn.click(
            fn=detect_brain_tumor,
            inputs=seg_inputs,
            outputs=seg_outputs,
        )

if __name__ == "__main__":
    demo.queue(max_size=10).launch(debug=True)