[HN Gopher] Vision Transformers Need Registers
___________________________________________________________________
Vision Transformers Need Registers
Author : felineflock
Score : 67 points
Date : 2025-04-28 16:53 UTC (6 hours ago)
(HTM) web link (arxiv.org)
(TXT) w3m dump (arxiv.org)
| refulgentis wrote:
| [2023]
|
| Previously:
|
| 2 years ago: https://news.ycombinator.com/item?id=37794996.
|
| 1 year ago: https://news.ycombinator.com/item?id=40329675
| bigdict wrote:
| Has this been used widely since?
| heyitsguay wrote:
| I ran a comparison of DINOv2 with and without registers on some
| image embedding tasks for work; DINOv2+registers saw a
| performance metric bump of 2-3%. Not nothing, not
| transformative, worth using when the only difference for
| inference is the model name string you're loading.
| ashvardanian wrote:
| I can't speak for the whole industry, but we used it in older
| UForm <https://github.com/unum-cloud/uform> and saw good
| adoption, especially among those deploying on the Edge, where
| every little trick counts. It's hard to pin down exact numbers
| since most deployments didn't go through Hugging Face, but at
| the time, these models were likely among the more widely
| deployed by device count.
| godelski wrote:
| yes
| kombine wrote:
| For example, it is used here
| https://github.com/facebookresearch/vggt/
| Scene_Cast2 wrote:
| So basically multiple CLS tokens.
|
| Fwiw, I tried multiple global tokens in my chess neural net and
| didn't see any uplift compared to my baseline of just having one.
| ninja3925 wrote:
| Note that it's not done for performance reason but rather to
| generate clear feature maps.
___________________________________________________________________
(page generated 2025-04-28 23:00 UTC)