ctuning/image-classification-tf-cpp.debian-9 repository overview

This container may be outdated. Please check CK-MLPerf docs⁠.

This repository contains Docker-ized instructions for the Collective Knowledge-powered image classification TFLite workloads in the MLPerf Inference benchmark: MobileNet (non-quantized, quantized), ResNet.

After pulling the latest image:

$ docker pull ctuning/image-classification-tf-cpp.debian-9

you can use it e.g. as follows:

• run image classification using non-quantized MobileNet on a single image:

$ docker run --rm ctuning/image-classification-tf-cpp.debian-9 \
"ck run program:image-classification-tf-cpp --dep_add_tags.weights=mobilenet,non-quantized"

• run image classification using ResNet on 10 images:

$ docker run --rm ctuning/image-classification-tf-cpp.debian-9 \
"ck run program:image-classification-tf-cpp --dep_add_tags.weights=resnet --env.CK_BATCH_COUNT=10"

• step into bash:

$ docker run -it --rm ctuning/image-classification-tf-cpp.debian-9 bash

For further information, see:

• build and run instructions⁠;
• benchmarking instructions⁠.