tff.simulation.baselines.cifar100.create_image_classification_task

Creates a baseline task for image classification on CIFAR-100.

The goal of the task is to minimize the sparse categorical crossentropy between the output labels of the model and the true label of the image.

train_client_spec A tff.simulation.baselines.ClientSpec specifying how to preprocess train client data.
eval_client_spec An optional tff.simulation.baselines.ClientSpec specifying how to preprocess evaluation client data. If set to None, the evaluation datasets will use a batch size of 64 with no extra preprocessing.
model_id A string identifier for a digit recognition model. Must be one of resnet18, resnet34, resnet50, resnet101 and resnet152. These correspond to various ResNet architectures. Unlike standard ResNet architectures though, the batch normalization layers are replaced with group normalization. </td> </tr><tr> <td>crop_height</td> <td> An integer specifying the desired height for cropping images. Must be between 1 and 32 (the height of uncropped CIFAR-100 images). By default, this is set to <a href="../../../../tff/simulation/baselines/cifar100#DEFAULT_CROP_HEIGHT"><code>tff.simulation.baselines.cifar100.DEFAULT_CROP_HEIGHT</code></a>. </td> </tr><tr> <td>crop_width</td> <td> An integer specifying the desired width for cropping images. Must be between 1 and 32 (the width of uncropped CIFAR-100 images). By default this is set to <a href="../../../../tff/simulation/baselines/cifar100#DEFAULT_CROP_WIDTH"><code>tff.simulation.baselines.cifar100.DEFAULT_CROP_WIDTH</code></a>. </td> </tr><tr> <td>cache_dir</td> <td> An optional directory to cache the downloadeded datasets. IfNone, they will be cached to~/.tff/. </td> </tr><tr> <td>use_synthetic_data` A boolean indicating whether to use synthetic CIFAR-100 data. This option should only be used for testing purposes, in order to avoid downloading the entire CIFAR-100 dataset.

A tff.simulation.baselines.BaselineTask.