Work

ImageNet 256x256 Log Densities for Learning Omega-Powered Conditional Distributions

Year: 2026

Type: dataset

Abstract: Aligned ImageNet class IDs, horizontal-flip flags, source-path metadata, and conditional and unconditional SiT log densities for 2,562,334 processed rows across 1,000 classes. Each source image is represented by original and horizontally flipped variants. Raw ImageNet pixels and VAE latent parameters are not included. This dataset accompanies the TMLR paper “Learning ω-Powered Conditional Distributions with Importance-Weighted Flow Matching.”

Source: Zenodo (CERN European Organization for Nuclear Research)

DOI: 10.5281/zenodo.23239356

Authors: Falkiewicz, Maciej, Lavda, Frantzeska, Ramapuram, Jason, Kalousis, Alexandros

Language: English

Topic: Generative Adversarial Networks and Image Synthesis

Subfield: Computer Vision and Pattern Recognition

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

OpenAlex ID: W7220948822

Cites:

Cited by:

Related to: 10