YOLOv8 Logistics Objects in Context (LOCO)

After trying out YOLOv8 with the salmon tracker, I decided to make a proof of concept for something that is more applicable to my work. For my job, I write CV algorithms that measure utilization of military warehouses collected from LiDAR devices. I decided to go on the hunt for some logistics datasets to do some object detection and tracking. I came across a Github page for a dataset called Logistics Objects in Context (LOCO). The dataset has object detection labels for forklifts, pallets, pallet trucks, and more. This dataset was labeled using COCO format. 

The Github page says the dataset is free to use, so go check it out and use it if you want! https://github.com/tum-fml/loco

Roboflow

I used Roboflow as a way to convert the annotations from COCO to YOLOv8 format. I hosted my dataset in their 'Universe'. Unless you are paying to use Roboflow, any dataset you make will be open to the public on the ‘Universe’. I did all my training on my PC, but I even trained a model on their website by using one of my 3 free credits. I show examples in the Google Colab notebook of how to access that via API instead of downloading and uploading a .pt file

https://universe.roboflow.com/tommy-sugg-uhvch/logistics-objects-in-context/dataset/1

Google Colab Notebook

You can get to the Colab notebook through my Github page. I provide my model and also link to the model hosted by Roboflow. I explain the packages that I am using in the notebook. I use both supervision and ByteTrack. ByteTrack has to be built from source. Both of them read from the ultralytics Model() class. I use ByteTrack for tracking detections and supervision for detections and visualization. Below is an example of what the outputs look like.