Saved in:
Bibliographic Details
Main Authors: Yue, Kaiyu, Jia, Menglin, Hou, Ji, Goldstein, Tom
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2602.15030
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914333186850816
author Yue, Kaiyu
Jia, Menglin
Hou, Ji
Goldstein, Tom
author_facet Yue, Kaiyu
Jia, Menglin
Hou, Ji
Goldstein, Tom
contents We introduce the Sphere Encoder, an efficient generative framework capable of producing images in a single forward pass and competing with many-step diffusion models using fewer than five steps. Our approach works by learning an encoder that maps natural images uniformly onto a spherical latent space, and a decoder that maps random latent vectors back to the image space. Trained solely through image reconstruction losses, the model generates an image by simply decoding a random point on the sphere. Our architecture naturally supports conditional generation, and looping the encoder/decoder a few times can further enhance image quality. Across several datasets, the sphere encoder approach yields performance competitive with state of the art diffusions, but with a small fraction of the inference cost. Project page is available at https://sphere-encoder.github.io .
format Preprint
id arxiv_https___arxiv_org_abs_2602_15030
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Image Generation with a Sphere Encoder
Yue, Kaiyu
Jia, Menglin
Hou, Ji
Goldstein, Tom
Computer Vision and Pattern Recognition
We introduce the Sphere Encoder, an efficient generative framework capable of producing images in a single forward pass and competing with many-step diffusion models using fewer than five steps. Our approach works by learning an encoder that maps natural images uniformly onto a spherical latent space, and a decoder that maps random latent vectors back to the image space. Trained solely through image reconstruction losses, the model generates an image by simply decoding a random point on the sphere. Our architecture naturally supports conditional generation, and looping the encoder/decoder a few times can further enhance image quality. Across several datasets, the sphere encoder approach yields performance competitive with state of the art diffusions, but with a small fraction of the inference cost. Project page is available at https://sphere-encoder.github.io .
title Image Generation with a Sphere Encoder
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2602.15030