Su Wang
@wangsu_gdm
Followers
28
Following
15
Media
0
Statuses
7
Senior Research Eng at Google DeepMind
Austin, TX
Joined December 2014
You heard it right: I obsessively took over 15k pictures over a multiyear period, worked with an amazing team to get them cleaned up, vetted, and captioned, and donated them for research. Hoping others will donate images in similar manner in future!
Image captions often lack the detail required to assess VL models. DOCCI descriptions are highly compositional, clearly delineating similarities and contrasts across images taken by @jasonbaldridge! Check our data visualizer: https://t.co/nm1QBg7U9k
13
35
226
Our work on text-image alignment evaluation has been accepted to ICLR 2024. Congrats to all our amazing coauthors, especially to our incredible project driver Jaemin @jmin__cho.
🚨New T2I Evaluation!🚨 We introduce Davidsonian Scene Graph (DSG) for reliable T2I evaluation with questions that: - are atomic and unique - cover full text prompt semantics (w/o hallucination) - and have valid consistencies https://t.co/JRLMvihjDT
@GoogleAI @uncnlp @uwnlp 🧵
0
6
25
Student researcher position applications are open at Google Deepmind! I'm hosting a SR in the intersection of bias and generative models. If you're an interested PhD student please reach out! https://t.co/dKPbGByGEb…
google.com
Find your next job at Google — Careers at Google.
0
20
48
Special thanks for the amazing foundational work done by the PaLM / PaLM 2 folks — automating complex text-image alignment question sets is a rough task, and PaLM truly shined. @YiTayML @_jasonwei @hwchung27 @iamandrewdai (and many others who aren’t active on twitter).
🚨New T2I Evaluation!🚨 We introduce Davidsonian Scene Graph (DSG) for reliable T2I evaluation with questions that: - are atomic and unique - cover full text prompt semantics (w/o hallucination) - and have valid consistencies https://t.co/JRLMvihjDT
@GoogleAI @uncnlp @uwnlp 🧵
0
0
4
🚨New T2I Evaluation!🚨 We introduce Davidsonian Scene Graph (DSG) for reliable T2I evaluation with questions that: - are atomic and unique - cover full text prompt semantics (w/o hallucination) - and have valid consistencies https://t.co/JRLMvihjDT
@GoogleAI @uncnlp @uwnlp 🧵
1
59
146
Baby boomers did a pretty good job teaching their millennial daughters that they could be anything they wanted to be and a pretty terrible job of preparing their sons for what that would mean for them as husbands and fathers
2K
21K
193K
New open-source language model from Google AI: Flan-T5 🍮 Flan-T5 is instruction-finetuned on 1,800+ language tasks, leading to dramatically improved prompting and multi-step reasoning abilities. Public models: https://t.co/bnYVnocJW2 Paper: https://t.co/3KPGJ3tgMw
37
469
2K