RAEv2 Data Pre processed datasets and pretrained encoders for RAEv2: Improved Baselines with Representation Autoencoders. All rights to the original owners; per subset attribution below. Repo Structure Data Pre processed datasets at 256x256. All rights to the original owners. Subset Task Source Format Notes imagenet 256 ImageNet ImageNet Arrow Or use your own ImageNet blip3o 256 T2I BLIP3o WDS Captioned image pairs render text 256 T2I RenderedText WDS Rendered text images scale rae 256 T2I Scale RAE WDS Synthetic FLUX images recon 256 NWM RECON WDS Robot navigation frames Pretrained Models Pretrained vision encoders and tokenizer weights used by RAEv2 (DINOv3, EUPE, iJEPA, MAE, MoCov3, SDVAE). All rights to the original owners. License & Attribution All rights belong to the original dataset and model owners listed above. This repository provides pre processed / packed versions of upstream content for efficient loading in RAEv2; upstream license terms apply to the underlying data. The packing layer itself is released under CC BY NC 4.0 per the RAEv2 codebase. Citation
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy