This repository contains trained YoloV5 models, as well as all tools necessary in order to have YoloV5 run on mobile.
We have several models which consider different datasets, i.e. recognize different objects:
- Bahnhof: this model recognizes the object typically found in a train station
- Wagen: this model recognizes the objects inside a train
- Traktion: this model is more specific, it handles all objects which have to do with locomotives
- Universal (all of the above)
The description of each label can be found here.
In order for a YoloV5 model to be used on mobile, one needs to convert it to a suitable format. This is CoreML for an iOS app and TFLite for an Android app. All information required to convert a model can be found here.
We have also developed libraries in order to facilitate the deployment of the converted YoloV5 model on mobile: Android library and iOS library
The Coco dataset format is a widely-used format for object detection. However YoloV5 does not use it and instead requires a custom structure. This converter allows to convert any Coco dataset into a Yolo dataset, which can then be used for training. All information required to convert a dataset can be found here.
The official yolov5 repository contains the script to train a YoloV5 model.
- Jeanne Fleury
- Stefan Aebischer
- Henrik Karppinen
Code released under the MIT.