In the Single Image Super Resolution with various prediction networks post, we used DIV2K and Flicker2K dataset for training. Below is a python snippet for both datasets. from glob import glob import torch from torch.utils.data import Dataset import torchvision from PIL import Image import numpy as np class SuperResolutionDataset(Dataset): def __init__(self, dataset_path, dataset_name_HR, dataset_name_LR, scale, name): self.name = name self.dataset_path = dataset_path self.dataset_name_HR = dataset_name_HR self.dataset_name_LR = dataset_name_LR self.scale = scale self....
The goal of single-image super-resolution (SISR) is to transform a low-resolution image into a high-resolution version, enhancing both its detail and quality. This process, called upscaling, increases the image鈥檚 spatial resolution while sharpening and clarifying its visual features. While the objective remains the same, the methods to achieve SISR can vary. In this post (Part 1), I鈥檒l explore three different neural network-based SISR approaches: color prediction, residual prediction, and kernel prediction....
In this post, I would like to introduce you to CoHiClust, a Contrastive Hierarchical Clustering model based on deep neural networks, which can be applied to typical image data. CoHiClust distills the base network into a binary tree without access to any labeled data. CoHiClust outperforms the established and popular Agglomerative Clustering and generates a hierarchical structure of clusters consistent with human intuition and image semantics. You can access the full paper of CoHiClust by clicking here....