BSD100HR

class deepinv.datasets.BSD100HR(root=None, download=False, transform=None, verbose=True, use_dict_output=False)

Bases: ImageFolder

Dataset for BSD100.

The BSD100 dataset [1] is a dataset consisting of 100 images commonly used for testing performance of image reconstruction algorithms. Images have sizes ranging from 240×160 to 480×320 pixels.

Raw data file structure:

self.root --- BSD100_HR.tar.gz
        |
        --- BSD100_HR --- 3096.png
        |               |
        |               --- 8023.png
        |               --- 12084.png
        |               --- ...
        |
        --- xxx

Raw dataset source : https://huggingface.co/datasets/eugenesiow/BSD100

Parameters:

• root (str) – Root directory of dataset. Directory path from where we load and save the dataset.

• download (bool) – If True, downloads the dataset from the internet and puts it in root directory. If dataset is already downloaded, it is not downloaded again. Default at False.

• transform: (Callable) – (optional) A function/transform that takes in a PIL image and returns a transformed version. E.g, torchvision.transforms.RandomCrop

• verbose (bool) – Print a message if the dataset has been correctly downloaded. Default True.

• use_dict_output (bool) – whether to return output as dict with keys “x”, “y”, “params” instead of tuple (default False).

---

References:

[1]

D. Martin, C. Fowlkes, D. Tal, and J. Malik. A database of human segmented natural images and its application to evaluating segmentation algorithms and measuring ecological statistics. In Proc. 8th Int’l Conf. Computer Vision, volume 2, 416–423. July 2001.

check_dataset_exists()

Verify that the image folders exist and contain all the images.

self.root should have the following structure:

self.root --- BSD100_HR --- 3096.png
        |               |
        |               --- 8023.png
        |               --- 12084.png
        |               --- ...
        |
        --- xxx