Menglin Jia

@menglin_jia

Research Scientist at Meta

kmnp.github.io

Joined October 2020

Replies Reposts Media

Menglin Jia
@menglin_jia
Oct 25, 2022
Attending #ECCV2022? Drop by the poster session 1.A (poster 084) about VPT! 🔥❄️ w/ @CornellCIS, @BelongieLab and @metaai !
Belongie Lab
@BelongieLab
Sep 23, 2022
1/3) What is the best way to adapt large pre-trained vision models to downstream tasks in terms of effectiveness and efficiency? Drawing inspiration from the recent advances on Prompting in NLP, we propose a new simple and efficient method: Visual Prompt Tuning (VPT) 👇
GIF
1
Menglin Jia
@menglin_jia
Oct 14, 2021
1/ Attending #ICCV2021? Drop by session 4 or ask us questions in the text Q&A about VisE, w/ @BelongieLab and @FacebookAI! Special thanks to Marseille who is featured in the teaser photo👇
Belongie Lab
@BelongieLab
Oct 12, 2021
1/ Can we leverage visual engagement as a supervisory signal? Join us at #ICCV2021 session 4 (10/12 Tue, 3-4 pm & 10/14 Thur, 8-9 am EDT) to discuss our recent work VisE with the aim to explore that question!
1
Menglin Jia
@menglin_jia
Jun 24, 2021
1/ Attending #CVPR2021? Come join the chat (session 10) or ask us questions in the text Q&A about Intentonomy, w/ @BelongieLab and @FacebookAI! Big thanks to my co-authors and @lukechesser @TimmyCarbone for providing the awesome @unsplash images!
Belongie Lab
@BelongieLab
Jun 24, 2021
1/ Why do we post images on social media platforms? Join us at #CVPR2021 session 10 (6/24 Thur, 11 am - 1:20 pm EDT) to discuss our recent work Intentonomy with the aim to explore that question!
00:00
2
Menglin Jia
@menglin_jia
May 31, 2021
📣 We show that the learned VisE models can outperform the ImageNet-supervised counterpart by a substantial margin in three subjective downstream tasks. Also shout-out to Marseille💕 for helping us demonstrate the idea 👇
Belongie Lab
@BelongieLab
May 31, 2021
(1/3) Introducing "VisE": a pretraining approach to learn visual representations via visual engagement signals. Given the same image, visual engagement provides semantically and contextually richer information than conventional recognition and captioning tasks.
00:00
Menglin Jia
@menglin_jia
May 27, 2021
Are objects and their image context sufficient for recognizing the intent behind images? Check out our recent work Intentonomy (to appear at @CVPR 2021).
Belongie Lab
@BelongieLab
May 27, 2021
(1/4) Understanding human intent behind images is particularly challenging, since it goes beyond standard visual recognition w/ @KMnP_menglin Zuxuan Wu @AustinReiter1 @clairecardie @SergeBelongie @sernamlim
00:00