predict.py

from PIL import Image
from yolo import YOLO
import gradio as gr
import os
from tqdm import tqdm

# Initialize YOLO model
yolo = YOLO()

def predict_image(image, crop=False, count=True):
    """
    Predict single image using YOLO model
    """
    try:
        r_image = yolo.detect_image(image, crop=crop, count=count)
        return r_image
    except Exception as e:
        print(f"Error: {e}")
        return None

def predict_directory(input_dir, output_dir, crop=False, count=True):
    """
    Predict images in a directory using YOLO model and save results to another directory
    """
    img_names = os.listdir(input_dir)
    results = []
    for img_name in tqdm(img_names):
        if img_name.lower().endswith(('.bmp', '.dib', '.png', '.jpg', '.jpeg', '.pbm', '.pgm', '.ppm', '.tif', '.tiff')):
            image_path = os.path.join(input_dir, img_name)
            image = Image.open(image_path)
            r_image = yolo.detect_image(image, crop=crop, count=count)
            if not os.path.exists(output_dir):
                os.makedirs(output_dir)
            output_path = os.path.join(output_dir, img_name.replace(".jpg", ".png"))
            r_image.save(output_path, quality=95, subsampling=0)
            results.append((img_name, output_path))
    return results

def inference(image, mode='predict', crop=False, count=True, input_dir=None, output_dir=None):
    if mode == 'predict':
        return predict_image(image, crop=crop, count=count)
    elif mode == 'dir_predict' and input_dir and output_dir:
        return predict_directory(input_dir, output_dir, crop=crop, count=count)
    else:
        raise ValueError("Invalid mode or missing directories for 'dir_predict' mode.")

title = "YOLO Image Prediction"
description = "This demo allows you to perform image prediction using a YOLO model. You can either predict a single image or all images in a directory."

css = """
    .image-frame img, .image-container img {
        width: auto;
        height: auto;
        max-width: none;
    }
"""

demo = gr.Interface(
    fn=inference,
    inputs=[
        gr.Image(type="pil", label="Input Image"),
        gr.Radio(choices=["predict", "dir_predict"], label="Mode", value="predict"),
        gr.Checkbox(value=False, label="Crop"),
        gr.Checkbox(value=True, label="Count"),
        gr.Textbox(placeholder="Input directory (for 'dir_predict' mode)", label="Input Directory", visible=False),
        gr.Textbox(placeholder="Output directory (for 'dir_predict' mode)", label="Output Directory", visible=False),
    ],
    outputs=gr.Image(type="pil", label="Output Image"),
    title=title,
    description=description,
    css=css,
)

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