sampler.py

"""
Copyright [2023] [Poutaraud]

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.

Adapted from 
https://github.com/sicara/easy-few-shot-learning/tree/master 
"""

import random
from typing import List, Tuple, Iterator

import torch
from torch import Tensor
from torch.utils.data import Sampler

class TaskSampler(Sampler):
    """
    Samples batches in the shape of few-shot image classification tasks. At each iteration, it will sample
    n_way classes, and then sample support and query images from these classes.
    """

    def __init__(
        self,
        dataset,
        n_way: int,
        n_shot: int,
        n_query: int,
        n_tasks: int,
    ):
        """
        Args:
            dataset: dataset from which to sample classification tasks. Must have a field 'label': a
                list of length len(dataset) containing containing the labels of all images.
            n_way: number of classes in one task
            n_shot: number of support images for each class in one task
            n_query: number of query images for each class in one task
            n_tasks: number of tasks to sample
        """
        super().__init__(data_source=None)
        self.n_way = n_way
        self.n_shot = n_shot
        self.n_query = n_query
        self.n_tasks = n_tasks

        self.items_per_label = {}
        for item, label in enumerate(dataset.__getlabel__()):
            if label in self.items_per_label.keys():
                self.items_per_label[label].append(item)
            else:
                self.items_per_label[label] = [item]

    def __len__(self) -> int:
        return self.n_tasks

    def __iter__(self) -> Iterator[List[int]]:
        for _ in range(self.n_tasks):
            yield torch.cat(
                [
                    # pylint: disable=not-callable
                    torch.tensor(
                        random.sample(
                            self.items_per_label[label], self.n_shot + self.n_query
                        )
                    )
                    # pylint: enable=not-callable
                    for label in random.sample(self.items_per_label.keys(), self.n_way)
                ]
            ).tolist()

    def episode(
        self, input_data: List[Tuple[Tensor, int]]
    ) -> Tuple[Tensor, Tensor, Tensor, Tensor, List[int]]:
        """
        Collate function to be used as argument for the collate_fn parameter of episodic data loaders.
        Args:
            input_data: each element is a tuple containing:
                - an image as a torch Tensor
                - the label of this image
        Returns:
            tuple(Tensor, Tensor, Tensor, Tensor, list[int]): respectively:
                - support images,
                - their labels,
                - query images,
                - their labels,
                - the dataset class ids of the class sampled in the episode
        """

        true_class_ids = list({x[1] for x in input_data})

        all_images = torch.cat([x[0].unsqueeze(0) for x in input_data])
        all_images = all_images.reshape((self.n_way, self.n_shot + self.n_query, *all_images.shape[1:]))

        support_images = all_images[:, :self.n_shot].reshape((-1, *all_images.shape[2:]))
        query_images = all_images[:, self.n_shot:].reshape((-1, *all_images.shape[2:]))

        all_labels = torch.tensor([true_class_ids.index(x[1]) for x in input_data])
        all_labels = all_labels.reshape((self.n_way, self.n_shot + self.n_query))

        support_labels = all_labels[:, :self.n_shot].flatten()
        query_labels = all_labels[:, self.n_shot:].flatten()

        return (support_images, support_labels, query_images, query_labels, true_class_ids)