INSANE! This TINY 2B AI Does SEVEN Vision Tasks on My RTX 4060 (WILD Results)
I tested Tencent's WeVisOne-VL (2B, run locally on an RTX 4060 laptop, plus the hosted 4B demo) on counting, dense detection, OCR, segmentation, matting and "find the thing that isn't there" prompts. It got most of them right and failed one. Counts from the two models differed, and I did not hand-count the birds image, so treat my numbers as a small hands-on test, not a benchmark. The 96.4% rejection figure is the authors' own, not mine. Chapters 0:00 The dog that isn't there 0:17 What is WeVisOne-VL? 0:28 Running it on an 8GB RTX 4060 0:36 Counting crowds and birds 1:00 Glass, text and cutout
Discussion
Rejoignez la communauté pour réagir et poser vos questions.
Aucun message pour l'instant. Lancez la discussion !