To train your model using mixed precision or TF32 with Tensor Cores or FP32, perform the following steps using the default parameters of the ResNext101-32x4d model on the ImageNet dataset. For the specifics concerning training and inference, see the Advanced section.

1. Clone the repository.

git clone https://github.com/NVIDIA/DeepLearningExamples
cd DeepLearningExamples/TensorFlow/Classification/ConvNets

2. Download and preprocess the dataset. The ResNext101-32x4d script operates on ImageNet 1k, a widely popular image classification dataset from the ILSVRC challenge.

• Download the images
• Extract the training and validation data:

mkdir train && mv ILSVRC2012_img_train.tar train/ && cd train
tar -xvf ILSVRC2012_img_train.tar && rm -f ILSVRC2012_img_train.tar
find . -name "*.tar" | while read NAME ; do mkdir -p "${NAME%.tar}"; tar -xvf "${NAME}" -C "${NAME%.tar}"; rm -f "${NAME}"; done
cd ..
mkdir val && mv ILSVRC2012_img_val.tar val/ && cd val && tar -xvf ILSVRC2012_img_val.tar

• Preprocess dataset to TFRecord form using script. Additional metadata from autors repository might be required.

3. Build the ResNext101-32x4d TensorFlow NGC container.

docker build . -t nvidia_rn50

4. Start an interactive session in the NGC container to run training/inference. After you build the container image, you can start an interactive CLI session with

nvidia-docker run --rm -it -v <path to imagenet>:/data/tfrecords --ipc=host nvidia_rn50

5. (Optional) Create index files to use DALI. To allow proper sharding in a multi-GPU environment, DALI has to create index files for the dataset. To create index files, run inside the container:

bash ./utils/dali_index.sh /data/tfrecords <index file store location>

Index files can be created once and then reused. It is highly recommended to save them into a persistent location.

6. Start training. To run training for a standard configuration (as described in Default configuration, DGX1V, DGX2V, single GPU, FP16, FP32, 90, and 250 epochs), run one of the scripts in the resnext101-32x4d/training directory. Ensure ImageNet is mounted in the /data/tfrecords directory.

For example, to train on DGX-1 for 90 epochs using AMP, run:

bash ./resnext101-32x4d/training/DGX1_RNxt101-32x4d_AMP_90E.sh /path/to/result /data

Additionally, features like DALI data preprocessing or TensorFlow XLA can be enabled with following arguments when running those scripts:

bash ./resnext101-32x4d/training/DGX1_RNxt101-32x4d_AMP_90E.sh /path/to/result /data --xla --dali

7. Start validation/evaluation. To evaluate the validation dataset located in /data/tfrecords, run main.py with --mode=evaluate. For example:

python main.py --arch=resnext101-32x4d --mode=evaluate --data_dir=/data/tfrecords --batch_size <batch size> --model_dir <model location> --results_dir <output location> [--xla] [--amp]

The optional --xla and --amp flags control XLA and AMP during evaluation.