44 KB Single Class Object Detector

Xailient has built the world’s smallest single class object detection computer vision model that can detect anything you train it on. It’s based on 9 years of research on bionic eyes and vision neuroscience. It’s 5000x smaller than YOLOv3 and can run on almost any IoT device since we have TensorFlow Lite, TensorFlow, Keras, and PyTorch versions of this model. Best of all, it is very robust and has been designed to generalize as you scale to several cameras. You can find some pre-trained models for use-cases like:

  • Face Detection
  • Car Detection
  • Person Detection
  • License Plate Reading

Here’s what the model output looks like (gives you both segmentation and bounding boxes):

And you can train custom models on your own data! Here’s a project where we trained the models to detect bees and Varroa mites inside bee hives using two cameras. All running on a Raspberry Pi 3B+ in real-time and powered by a solar panel in the middle of nowhere!

Head to Xailient to start training yours!

This is really cool @ShivyXailient thank you for sharing, what inference does the model run on? OpenVINO etc?

Thanks Paul. We’ve compiled the model to run on several chips within the Intel family (Core x86-64, Atom, Movidius NCS, MyriadX) using OpenVINO. And many other architectures (ARM, Coral, Android, iOS, Sony, etc).

Right now we only support ARM and x86-64 direct-downloadable SDKs from our website but the rest are coming based on demand. SDKs come in both Python and C++ flavours. If you want us to compile an SDK that’s not supported please email us and we’ll do our best to sort you out! Hope that helps!