PyTorch Hub and autoShape update

[glenn-jocher]

This PR pushes the autoShape concept further with a Detections() class produced by YOLOv5 inference now when using .autoShape(). PyTorch Hub tutorial updated also, and we are also finalizing an official submission to https://github.com/pytorch/hub.

Everything works super simply. To see a preview simply run $ python hubconf.py. Example:

import torch
from PIL import Image

# Model
model = torch.hub.load('ultralytics/yolov5', 'yolov5s', pretrained=True).autoshape()  # for PIL/cv2/np inputs and NMS

# Images
img = Image.open('zidane.jpg')  # PIL image

# Inference
results = model(img)  # includes NMS

# View
results.show()  # .show() results, .save() jpgs, or .print() to screen

# Data
results.xyxy  # x1, y1, x2, y2 (pixels) format
results.xyxyn  # x1, y1, x2, y2 (normalized) format
results.xywh  # x_center, y_center, width, height (pixels) format
results.xywhn  # x_center, y_center, width, height (normalized) format

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Enhancements to YOLOv5 inference and image processing.

📊 Key Changes

• 🧹 Simplified conditional checks (e.g., if len(det) instead of if det is not None and len(det)).
• 🎨 Integrated the Pillow (PIL) library for image handling, expanding dependency list to include pillow.
• 🖼️ Modified autoShape and Detections classes to improve image input handling, normalization, and post-processing.
• 🚀 Introduced a new Detections class to encapsulate detection results with various output functions (print, show, save).
• 🛠️ Minor code refactors and consistency improvements (e.g., replaced cloning tensors instead of creating new ones for shape manipulations).

🎯 Purpose & Impact

• 📈 Improved code readability and maintainability by removing unnecessary checks and clarifying comments.
• ✨ Pillow integration aims to enhance image operations and make the library more user-friendly for image-related tasks.
• 🌐 The new Detections class provides a standard way to interact with inference outputs, easing the display and interpretation of results for users.
• 🏗️ Refactoring contributes to more consistent and efficient codebase, potentially leading to fewer bugs and better performance.
• Overall, these changes should provide a smoother user experience and facilitate easier integration and use of YOLOv5 in image detection tasks.