import gradio as gr import sahi import torch from ultralyticsplus import YOLO, render_model_output # Download sample images sahi.utils.file.download_from_url( "https://raw.githubusercontent.com/kadirnar/dethub/main/data/images/highway.jpg", "highway.jpg", ) sahi.utils.file.download_from_url( "https://raw.githubusercontent.com/obss/sahi/main/tests/data/small-vehicles1.jpeg", "small-vehicles1.jpeg", ) sahi.utils.file.download_from_url( "https://raw.githubusercontent.com/ultralytics/yolov5/master/data/images/zidane.jpg", "zidane.jpg", ) # List of YOLOv8 segmentation models model_names = [ "yolov8n-seg.pt", "yolov8s-seg.pt", "yolov8m-seg.pt", "yolov8l-seg.pt", "yolov8x-seg.pt", ] current_model_name = "yolov8m-seg.pt" model = YOLO(current_model_name) def yolov8_inference( image: gr.Image = None, model_name: gr.Dropdown = None, image_size: gr.Slider = 640, conf_threshold: gr.Slider = 0.25, iou_threshold: gr.Slider = 0.45, ): """ YOLOv8 inference function to return masks and label names for each detected object Args: image: Input image model_name: Name of the model image_size: Image size conf_threshold: Confidence threshold iou_threshold: IOU threshold Returns: Object masks, coordinates, and label names """ global model global current_model_name # Check if a new model is selected if model_name != current_model_name: model = YOLO(model_name) current_model_name = model_name # Set the confidence and IOU thresholds model.overrides["conf"] = conf_threshold model.overrides["iou"] = iou_threshold # Perform model prediction results = model.predict(image, imgsz=image_size, return_outputs=True) # Initialize an empty list to store the output output = [] # Iterate over the results for result in results: # Check if segmentation masks are available if 'masks' in result and result['masks'] is not None: masks = result['masks']['data'] for i, (mask, box) in enumerate(zip(masks, result['boxes'])): label = model.names[int(result['boxes']['cls'][i])] mask_coords = mask.tolist() # Convert mask coordinates to list format output.append({"label": label, "mask_coords": mask_coords}) else: # If masks are not available, just extract bounding box information for i, box in enumerate(result['boxes']): label = model.names[int(result['boxes']['cls'][i])] bbox = box['xyxy'].tolist() # Bounding box coordinates output.append({"label": label, "bbox_coords": bbox}) return output # Define Gradio interface inputs and outputs inputs = [ gr.Image(type="filepath", label="Input Image"), gr.Dropdown( model_names, value=current_model_name, label="Model type", ), gr.Slider(minimum=320, maximum=1280, value=640, step=32, label="Image Size"), gr.Slider( minimum=0.0, maximum=1.0, value=0.25, step=0.05, label="Confidence Threshold" ), gr.Slider(minimum=0.0, maximum=1.0, value=0.45, step=0.05, label="IOU Threshold"), ] # Output is a dictionary containing label names and coordinates of masks or boxes outputs = gr.JSON(label="Output Masks and Labels") title = "Ultralytics YOLOv8 Segmentation Demo" # Example images for the interface examples = [ ["zidane.jpg", "yolov8m-seg.pt", 640, 0.6, 0.45], ["highway.jpg", "yolov8m-seg.pt", 640, 0.25, 0.45], ["small-vehicles1.jpeg", "yolov8m-seg.pt", 640, 0.25, 0.45], ] # Build the Gradio demo app demo_app = gr.Interface( fn=yolov8_inference, inputs=inputs, outputs=outputs, title=title, examples=examples, cache_examples=False, # Set to False to avoid caching issues theme="default", ) # Launch the app demo_app.queue().launch(debug=True)