HackerTrans
TopNewTrendsCommentsPastAskShowJobs

globalvisualmem

no profile record

Submissions

Show HN: SuperVisual platform for crowdsourcing dataset and action transformers

supervisual.app
2 points·by globalvisualmem·4 ปีที่แล้ว·0 comments

Show HN: Use Screen Sharing to capture anything, search with CLIP in JavaScript

supervisual.app
2 points·by globalvisualmem·4 ปีที่แล้ว·0 comments

Show HN: Private AI visual search with Screen Recording in Chrome browser tab

supervisual.app
2 points·by globalvisualmem·4 ปีที่แล้ว·1 comments

comments

globalvisualmem
·4 ปีที่แล้ว·discuss
- Scraping is a major issue.

- Browser privacy / VPNs lead to an arms race where account. age/reputation become critical signal.

- Apple/Google play store apps provide device integrity check which improves confidence in account reputation / Ad targeting.

- Everyone already has a Google/Apple account and are unlikely to convert if they wont just signup via SSO which takes couple of clicks.
globalvisualmem
·4 ปีที่แล้ว·discuss
I just tried it works beautifully! I suggest defaulting to lower resolution 384 X 384 since it will speed up.
globalvisualmem
·4 ปีที่แล้ว·discuss
SuperVisual uses latest advances in AI (CLIP) along with features in Chrome/Edge browsers to provide a private visual search experience over any visual content using screen/tab sharing feature (GetDisplayMedia). SuperVisual runs entirely in the browser, without requiring any extension. It's made possible thanks to TensorFlow.js, ONNX JS runtime and support for novel WebRTC / Media features such as WebCodecs, Tab sharing & Capture handle. You can see more video demos in this playlist.

https://www.youtube.com/watch?v=A_CcS-KWWNY&list=PLBo8nYvYvq...
globalvisualmem
·4 ปีที่แล้ว·discuss
There is also recent work by Google called DreamBooth, though similar to Imagen/Parti Google refuses to release any model or code.

https://dreambooth.github.io/