29 datasets found

The Vaihingen dataset consists of 1440 scenes with a size of 250×250 pixels. Each scene is a colour-infrared (CIR) true orthophoto and a height grid (digital surface model; DSM)...

The Synthia dataset is a large-scale urban scene understanding dataset, containing 9000 samples. It is used for semantic segmentation tasks.

Few-shot segmentation remains challenging due to the limitations of its labeling information for unseen classes. Most previous approaches rely on extracting high-level feature...

The GTA dataset consists of 24966 synthetic images synthetically generated from a video game consisting of outdoor scenes with rich variety of variations in lighting and traffic...

The NYUD-v2 dataset is a benchmark for indoor scene segmentation and depth estimation. It contains 1449 images with 4 tasks: semantic segmentation, depth estimation, surface...

The Synthia→Cityscapes task is a domain adaptation task for semantic segmentation, where the source domain is Synthia and the target domain is Cityscapes.

The dataset used in the paper is GTA5 and SYNTHIA, which are used for domain adaptive semantic segmentation (DASS).

The ADE20K dataset is a large-scale dataset for semantic segmentation. It contains 20,000 images with 150 semantic categories, with 20,000 images for training, 2,000 images for...

CamVid dataset is a benchmark dataset for semantic segmentation. It consists of 700 images with 11 object classes.

The dataset used in the paper is the Pascal VOC 2012 dataset, which is a benchmark for instance segmentation. The dataset consists of 1464 images with 20 class categories and...

COCO Stuff dataset is an extension of the COCO dataset, 164,000 images covering 171 classes are annotated with segmentation masks.

Pyramid scene parsing network for semantic segmentation.

The PASCAL Context dataset is a benchmark for multi-task learning in computer vision. It contains 10103 images with 5 tasks: semantic segmentation, human body part segmentation,...

The dataset used in the paper is a pre-trained ResNet-50 classifier, which is used for image synthesis, unpaired image-to-image translation, and feature similarity estimation.

CelebAMask-HQ provides the parsing map of images in CelebA-HQ down-sampled to 512 × 512, where pixel-level annotation of 19 classes, including facial components and accessories,...

The dataset is used for indoor scene understanding and contains RGB and depth images.

Multi-label image recognition is a practical and challenging task compared to single-label image classification.

Multi-task learning (MTL) research is broadly divided into two categories: one is to learn the correlation between tasks through model structures, and the other is to balance...

The dataset is generated using a latent diffusion model, specifically Stable Diffusion 2.1, and is used for semantic segmentation tasks.

The LoveDA dataset contains high-spatial-resolution images from three different cities, focusing on improving the generalization capability of model from different urban and...