Hacky Hour 23: Open Office Hour 1/21/21

Hacky Hour: Open Office Hour

Open Office Hour

This week at Hacky Hour, alwaysAI’s CTO, Steve Griset, and Developer Advocate and Software Engineer, Lila Mullany hosted alwaysAI’s Open Office Hour. This Hacky Hour format is perfect for both beginners and experienced alwaysAI developers. We are here to ensure the success of our community. In these sessions, our community can ask us questions about setting up an edge device, building a CV application, or general inquiries related to computer vision and alwaysAI!

Guest Questions

QUESTION: What types of models can we train with the alwaysAI CLI? What is the output format of alwaysAI’s custom trained models?

ANSWER (Lila): You can train object detection models. There are two frameworks developers can train from, namely, mobilenet_ssd and Faster R-CNN.

(Steve): We don’t recommend deploying Faster R-CNN on an edge device because these tend to be slower. You can also contact us to convert your model to TensorRT format.

QUESTION: Why might someone want to convert a model into the TensorRT format from mobilnet_ssd?

ANSWER (Steve): To get real-time accuracy on a CV application, we recommend converting the model to a TensorRT model to accelerate the performance. See below for alwaysAI’s performance testing numbers.

performance testing chart

QUESTION: Does alwaysAI support the Google Coral Accelerator for training and inferencing?

ANSWER (Steve): Not at this time. We support Myriad devices and NVIDIA Jetson devices.

QUESTION: What is the recommendation to improve accuracy for a CV model?

ANSWER (Todd): The first step to improving the accuracy of the model is improving the dataset. The best way to do this is to evaluate your production environment and iterate the dataset through the model training process to bet fit the given production environment.

QUESTION: Will alwaysAI support Keras?

Answer (Steve): Keras is a model retraining tool. We are looking to support models built on Keras. You can contact us on Discord to convert your model to TensorFlow.

QUESTION: Does the Jetson Nano support TensorRT?

Answer (Steve): Yes! It is an inference engine. You can use the Cuda Toolkit, which can use the CPU and GPU in parallel for processing.

QUESTION: Does edgeIQ support object re-identification?

Answer (Lila): Yes, we have a few centroid tracking, which associates a bounding box with an object ID based on distances from previous detections, and correlation-tracking algorithms, which tracks objects based on a correlation tracking algorithm. Learn more about tracking with alwaysAI.

QUESTION: Is it possible to retrain a Yolo model with alwaysAI?

Answer (Todd): Yes, as of March, it will be possible to train a Yolo model with the Model Training Toolkit.

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.

Get started now

We are providing professional developers with a simple and easy-to-use platform to build and deploy computer vision applications on edge devices. 

Sign Up for Free