Saved in:
| Main Authors: | , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2603.22387 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911557501321216 |
|---|---|
| author | Zhu, Chenchen Suri, Saksham Jose, Cijo Oquab, Maxime Szafraniec, Marc Wen, Wei Xiong, Yunyang Labatut, Patrick Bojanowski, Piotr Krishnamoorthi, Raghuraman Chandra, Vikas |
| author_facet | Zhu, Chenchen Suri, Saksham Jose, Cijo Oquab, Maxime Szafraniec, Marc Wen, Wei Xiong, Yunyang Labatut, Patrick Bojanowski, Piotr Krishnamoorthi, Raghuraman Chandra, Vikas |
| contents | Running AI models on smart edge devices can unlock versatile user experiences, but presents challenges due to limited compute and the need to handle multiple tasks simultaneously. This requires a vision encoder with small size but powerful and versatile representations. We present our method, Efficient Universal Perception Encoder (EUPE), which offers both inference efficiency and universally good representations for diverse downstream tasks. We achieve this by distilling from multiple domain-expert foundation vision encoders. Unlike previous agglomerative methods that directly scale down from multiple teachers to an efficient encoder, we demonstrate the importance of first scaling up to a large proxy teacher and then scaling down from this single teacher. Experiments show that EUPE achieves on-par or better performance than individual domain experts of the same size on diverse task domains and also outperforms previous agglomerative encoders. We release the full family of EUPE models and the code to foster future research. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_22387 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Efficient Universal Perception Encoder Zhu, Chenchen Suri, Saksham Jose, Cijo Oquab, Maxime Szafraniec, Marc Wen, Wei Xiong, Yunyang Labatut, Patrick Bojanowski, Piotr Krishnamoorthi, Raghuraman Chandra, Vikas Computer Vision and Pattern Recognition Running AI models on smart edge devices can unlock versatile user experiences, but presents challenges due to limited compute and the need to handle multiple tasks simultaneously. This requires a vision encoder with small size but powerful and versatile representations. We present our method, Efficient Universal Perception Encoder (EUPE), which offers both inference efficiency and universally good representations for diverse downstream tasks. We achieve this by distilling from multiple domain-expert foundation vision encoders. Unlike previous agglomerative methods that directly scale down from multiple teachers to an efficient encoder, we demonstrate the importance of first scaling up to a large proxy teacher and then scaling down from this single teacher. Experiments show that EUPE achieves on-par or better performance than individual domain experts of the same size on diverse task domains and also outperforms previous agglomerative encoders. We release the full family of EUPE models and the code to foster future research. |
| title | Efficient Universal Perception Encoder |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2603.22387 |