CelebA-HQ

Introduced by Karras et al. in Progressive Growing of GANs for Improved Quality, Stability, and Variation

The CelebA-HQ dataset is a high-quality version of CelebA that consists of 30,000 images at 1024×1024 resolution.

Source: IntroVAE: Introspective Variational Autoencoders for Photographic Image Synthesis

Homepage

Task	Dataset Variant	Best Model
Image Generation	CelebA-HQ 256x256	RDM
Image Generation	CelebA-HQ 1024x1024	RDM
Image Generation	CelebA-HQ 128x128	U-Net GAN
Blind Face Restoration	CelebA-HQ	GPEN
Image-to-Image Translation	CelebA-HQ	StarGAN v2
Image Inpainting	CelebA-HQ	MAT
Multimodal Unsupervised Image-To-Image Translation	CelebA-HQ	StarGAN v2
Image Generation	CelebA-HQ 64x64	VAEBM
Image Super-Resolution	CelebA-HQ 128x128	IDM
Unconditional Image Generation	CelebA-HQ	ConPreDiff
Density Estimation	CelebA-HQ 256x256	RNODE
Image Generation	CelebA-HQ 512x512	RDM
Image Super-Resolution	Celeb-HQ 4x upscaling	Edge-informed SR