dualvtable/idealo-superres-ngc-tf repository overview

This image is based off the official Docker image provided by the Idealo Image Super-Resolution (ISR) project⁠.

The original image from Idealo uses the upstream TensorFlow 1.13.1-gpu-py3 image and it doesn't have great prediction performance compared to the NVIDIA NGC TensorFlow container (on several workloads I tried, I was able to get 50% better performance).

So, we will switch to building an image off the latest NGC TensorFlow container: nvcr.io/nvidia/tensorflow:21.02-tf1-py3.

The Dockerfile is provided for informational purposes. You would need to override the paths to the datasets (data), pre-trained weights for the ISR models (weights) and the configuration file (config.yml) that controls the parameters.

# $ sudo docker run --gpus all -it \
#   --shm-size=1g --ulimit memlock=-1 --ulimit stack=67108864 \
#   -v /tmp/data:/workspace/superres/data \
#   -v /tmp/weights:/workspace/superres/weights \
#   -v /tmp/config.yml:/workspace/config.yml \
#   dualvtable/idealo-superres-ngc-tf:21.02-py3
FROM nvcr.io/nvidia/tensorflow:21.02-tf1-py3

COPY ISR ./ISR
COPY config.yml ./

RUN pip install --upgrade pip
RUN pip install imageio

ENV PYTHONPATH ./ISR/:$PYTHONPATH
ENTRYPOINT ["python", "ISR/assistant.py", "--prediction", "--default", "--config"]
CMD ["config.yml"]