For detailed instructions on getting started with TensorFlow with DirectML, see the tensorflow-directml GitHub repo and GPU accelerated ML training.
Three sample models are included in this repo to help you get started. These samples include both inference and training scripts, and you can either train the models from scratch or use the supplied pre-trained weights.
For users of the TensorFlow-DirectML package forked from TF 1.15, the TF1 versions of the models can be found below.
- squeezenet - a small image classification model
- VisionTransformer - a transformer-based image classification model
- yolov3 - real-time object detection model
For users of the TensorFlow-DirectML plugin package on TF 2.9+, the TF2 versions of the models can be found below.