Hacky Hour 29: Model Training with YOLO Models

By Komal Devjani • Mar 09, 2021

Model Training with YOLO Models

What are YOLO models? 

The “You Only Look Once,” or YOLO, family of models are a series of end-to-end deep learning models designed for fast object detection, developed by Joseph Redmon, et. al. Now you can train and customize YOLO models on the Model Training Toolkit with alwaysAI. In this Hacky Hour, Todd Gleed, Product Manager at alwaysAI, demonstrated how to customize YOLO models with the Model Training Toolkit for your use case.

Guest Questions

QUESTION: Can we customize our labels?

ANSWER (Steve Bottos): Yes you can. You'll need to first create a dataset with your annotations, and when you select that dataset to train on it will automatically detect the labels and structure the model accordingly.

QUESTION: I have worked with YOLO. How would you compare this version with that of version 4 and 5?

ANSWER (Steve Bottos): The difference between YOLO v4 and v5 is architectural. There is an accuracy and speed boost between the two; Yolo v4 being superior. YOLO v5 is actually a reimplementation of YOLO v4, its not canonical. YOLO v5 is created with Ultralytics, and it was converted to PyTorch making it easy to deploy. YOLO v5 performs worse on inference time and accuracy than v4. There will be no issues with deploying YOLO v4 with alwaysAI, however we do not yet support YOLO v5. We also recommend using YOLO v4 as opposed to YOLO v3.

QUESTION: If the training failed, can I restart the training at the last checkpoint since the version is not available for the new model? Both for command line and notebook GUI

ANSWER (Todd): If there is an error, we do save checkpoints throughout the training process. If you manually stop it on the desktop, then you can retrieve that. If something unknown happens, we can use the saved checkpoints to save the trained model. Reach out to us if you’re having trouble retrieving your model after an error. 

QUESTION: How did you decide to make 320x320 the default? 

ANSWER (Todd): The reason for that is that it's as small as you can go with YOLO models. We stuck with the lowest as the default because when you run this on the edge device, you don’t want to add any additional overhead.


See below for the full video of the Hacky Hour, or click here.

Open Office Hour

Join us every Thursday at 10:30 AM PST for weekly Hacky Hour! Whether you are new to the community or an experienced user of alwaysAI, you are welcome to join, ask questions, and provide the community with information about what you're working on. Register here.


By Komal Devjani • Mar 09, 2021

Developer stories to your inbox.

Subscribe to the Developer Digest, a monthly dose of all things code.

You may unsubscribe at any time using the unsubscribe link in the digest email. See our privacy policy for more information.

alwaysAI Ad
stylized image of a computer chip

Sign up today and start your project

We can't wait to see what you'll build!