You Always Log Everything (YALE)

Logging YOLOv5 Experiments with W&B

YOLO ("You Only Look Once") provides tools for real-time object detection with convolutional neural networks.

YOLO now works with Weights & Biases, an experiment tracking toolkit, so you can keep track of all the hyperparameters you've tried, view real-time updates on system and model metrics, version and store datasets and models, and more!

In this colab, we'll show you how to use YOLO and W&B together. It's as easy as running a single pip install before you run your YOLO experiments!

Follow along with a video tutorial on YouTube.

0. Setup

First, let's get ourselves organized: clone the repo, install our dependencies, and confirm we've got PyTorch and a GPU.

1. Inference

Now, let's apply a pre-trained, already existing object detection network.

detect.py runs inference on a variety of sources, downloading models automatically from the latest YOLOv5 release.

Here, we'll just run a sample image through that network to make sure everything is working.

Results are saved to runs/detect. A full list of available inference sources:

2. Training to Fine-Tune

For applications, it's often important to take a pre-trained model and fine-tune it to work on a specific dataset -- for example, in a construction safety application, we might use fine-tuning to specialize our network in detecting the presence/absence of protective equipment.

We'll mimic this process on the COCO128 image tutorial dataset.

Weights & Biases Logging (🚀 NEW)

Weights & Biases (W&B) is now integrated with YOLOv5 for real-time visualization and cloud logging of training runs. This allows for better run comparison and introspection, as well improved visibility and collaboration among team members. To enable W&B logging install wandb, and then train normally (you will be guided through setting up wandb account during your first use).

And that's it! So long as W&B is installed, you'll get rich, detailed metrics in a live dashboard accessible from a browser on any device.

Just click the link that appears below next to wandb and the 🚀 emoji.

With W&B, during training you will see live updates on the dashboard at wandb.ai, including interactive bounding box visualizations (look for a panel called "Images" in the Media panel section), and you can create and share detailed Reports of your results. For more information see the YOLOv5 Weights & Biases Tutorial or check out the video tutorial for this notebook.