Dataset Hub

The AdvancedEdit dataset was created via a novel data construction pipeline, featuring high visual quality, complex instructions, and good background consistency at a large scale.

This dataset includes three features: instruction, input, and output, all of type string. The dataset contains only a training split (train) with 6,856 samples, total size 1,958,991 bytes. Download size is 792,005 bytes. In the default configuration, the data file path is data/train-*.

The CIFAR-100-Enriched dataset is an augmented version of the original CIFAR-100 dataset, comprising 60,000 color images of 32 × 32 pixels across 100 classes, with 600 images per class. In addition to fine‑grained labels (specific categories), it includes coarse‑grained labels (super‑categories). Augmentation includes image embeddings generated by a Vision Transformer, facilitating data analysis and model training. The dataset aims to promote data‑driven AI principles and advance research in image classification.

GarVerseLOD is a high‑fidelity 3D garment reconstruction dataset created by The Chinese University of Hong Kong (Shenzhen). It contains 6,000 garment models handcrafted by professional artists with fine geometric details. The dataset offers three Levels of Detail (LOD): a coarse shape with no detail, a stylized shape with pose‑blended detail, and a pixel‑aligned detail level. During creation, a conditional diffusion model generated a large number of high‑quality paired images to enhance dataset generalization. GarVerseLOD is primarily intended for single‑image in‑the‑wild 3D garment reconstruction, addressing the limitations of existing methods in handling complex garment deformations and diverse poses.