Standard Vision Model
YoloX is the best way to do Computer Vision. And Sentigral Vision is the easiest way to do YoloX.
The very easiest way of all is to use our default container. It’s available on AWS Marketplace, installs immediately, and identifies 80 classes of objects, because it’s based on the COCO dataset. Here’s the full list of objects it can detect, and here’s some more information on COCO.
In one container, you can choose which of the six sizes of YoloX to use – X, L, M, S, Tiny or Nano – so you can get the perfect balance of accuracy and speed for any task. The sample pictures to the right have been processed with YoloX-X.
For most tasks, Nano is plenty, but if you want to track more than about 32 objects simultaneously, or if bounding box precision is very important, you may want a larger version. You can send any of your data to any of the models in the container at any time.
Simply download the documentation on how to do that, and everything else.
If you need any help or want to chat about having a custom model built, open a Live Chat with us using the balloon in the bottom right.
YoloX-X model
The images above use our YoloX-X model.
Model size – X and Nano
Below, you can compare the two extremes of model size – X and Nano – on a relatively complex scene. X catches more of the objects, has no false positives, and more accurate box corners. That said, Nano will still work fine for many tasks, and is extremely fast – around 9000 frames per second on a p4d.24xlarge instance, compared to 3000 frames per second using X.
A set containing hundreds of images and videos processed by all six sizes of our models is available to download from Google Drive and from Dropbox.