Erroll Wood

33 posts

Erroll Wood

Erroll Wood

@errollw

Staff SWE at Google, working on Digital Humans for AR & VR

Cambridge, UK Katılım Ocak 2011
207 Takip Edilen376 Takipçiler
Erroll Wood
Erroll Wood@errollw·
Meet D4RT 🎯 A unified model for next-gen scene reconstruction and tracking. 🧠 The Secret Sauce: Incredible results thanks in part to diverse, realistic synthetic data from our team. The results are wild. 🤯
Google DeepMind@GoogleDeepMind

We're helping AI to see the 3D world in motion as humans do. 🌐 Enter D4RT: a unified model that turns video into 4D representations faster than previous methods - enabling it to understand space and time. This is how it works 🧵

English
0
3
10
940
Erroll Wood
Erroll Wood@errollw·
It turns out that a synthetic-only face prior is great for novel view synthesis 🦾. If you're heading to SIGGRAPH Asia 🇯🇵, don't miss Cᴀꜰᴄᴀ. Great work, @mc_buehler!
Marcel C. Buehler@mc_buehler

A purely synthetic face prior can generalize to casual in-the-wild captures and stylized faces 🤯 I'm happy to share our @SIGGRAPHAsia paper “Cafca: High-quality Novel View Synthesis of Expressive Faces from Casual Few-shot Captures.” syntec-research.github.io/Cafca/

English
0
0
6
357
Erroll Wood
Erroll Wood@errollw·
Coming to #CVPR23? Come to my talk on Sunday morning at the GAZE workshop gazeworkshop.github.io/2023/ to hear about Synthetic Data for Gaze Estimation & More. Find out what worked 🙌🏻, what didn't 🤔, and what's next 🚀.
Erroll Wood tweet media
English
0
3
46
4.6K
Erroll Wood
Erroll Wood@errollw·
@skynetislov3 We hear you, and understand! We have nothing more to share at this point in time, but keep watching this space 🙂
English
0
0
0
0
gpu go brr...
gpu go brr...@skynetislov3·
@errollw If only there was code for this. As I am sure you know research projects with code are much more likely to be expanded upon and subsequently more frequently cited.
English
1
0
0
0
Erroll Wood
Erroll Wood@errollw·
@designerzen We have nothing more to share at this time, but keep watching this space 🙂
English
1
0
0
0
zen
zen@designerzen·
@errollw How exciting! Amazing work! Are there any online implementations or demos of these libraries? I'm very interested in using this in my face controlled musical synthesizer and this looks far superior to what I am currently using! So accurate and fast! 😍
English
1
0
1
0
Erroll Wood
Erroll Wood@errollw·
@cvtalks That's a really cool dataset! The community will love it! However, I'd argue that #syntheticdata can give you more consistent labels, and greater training data diversity.
English
0
0
0
0
Одна інтеллектуальна одиниця
Одна інтеллектуальна одиниця@cvtalks·
> While a human can consistently label face images with e.g., 68 landmarks, it would be almost impossible for them to annotate an image with dense landmarks. That's not fully correct. github.com/PinataFarms/DA… gives you dense 3D Heads annotations of real persons. You're welcome :)
Erroll Wood@errollw

Very proud to share more about our work on 𝗗𝗲𝗻𝘀𝗲 𝗙𝗮𝗰𝗲 𝗟𝗮𝗻𝗱𝗺𝗮𝗿𝗸𝘀, which will appear at #ECCV2022. Check it out! 👇 🔗 microsoft.github.io/DenseLandmarks/ 🎥 youtube.com/watch?v=M8NNyt… 📝 arxiv.org/abs/2204.02776

English
1
0
7
0
Erroll Wood
Erroll Wood@errollw·
@afromero I'm not sure, but please do reach out to the organizers of the workshop to find out :)
English
0
0
0
0
Andrés Romero
Andrés Romero@afromero·
@errollw Are you planning on recording the Workshop and uploading it to youtube?
English
1
0
1
0
Babak Taati
Babak Taati@Babak_Taati·
Journal club Thursday June 2nd at 2pm: We will be discussing "3D face reconstruction with dense landmarks" by @errollw et al. @KITE_UHN @KITETrainees @AGEWELL_NCE cs.toronto.edu/~taati/journal…
Erroll Wood@errollw

I'm delighted to share our work on dense landmarks! arxiv.org/abs/2204.02776 🌟 Dense landmarks are all you need for accurate 3D face reconstruction and performance capture 💪 No surprise: the secret sauce is synthetic data 🏃 It's fast, running at 150FPS on a single CPU thread

English
1
1
5
0
Erroll Wood
Erroll Wood@errollw·
@91amin91 Mediapipe is awesome, and I love their Attention Mesh design. But I think our synthetic training data is just better than their real data😉 We have nothing to share at this time unfortunately, but please stay tuned for further announcements🙂
English
0
0
0
0
This.is.Amin
This.is.Amin@91amin91·
@errollw Wow, congratulations,you left Mediapipe in dust! Is it going to be publicly available?
English
1
0
0
0
Erroll Wood
Erroll Wood@errollw·
I'm delighted to share our work on dense landmarks! arxiv.org/abs/2204.02776 🌟 Dense landmarks are all you need for accurate 3D face reconstruction and performance capture 💪 No surprise: the secret sauce is synthetic data 🏃 It's fast, running at 150FPS on a single CPU thread
Erroll Wood tweet media
English
14
68
411
0
Erroll Wood
Erroll Wood@errollw·
@MrCatid We have nothing to share at this time unfortunately, but please stay tuned for further announcements🙂
English
1
0
0
0
Erroll Wood
Erroll Wood@errollw·
@cubaneth We have nothing to share at this time unfortunately, but please stay tuned for further announcements🙂
English
0
0
1
0
Erroll Wood
Erroll Wood@errollw·
@Wiz64268099 We have nothing to share at this time unfortunately, but please stay tuned for further announcements🙂
English
1
0
1
0
Wiz
Wiz@mantravaadi·
@errollw opensource ?
English
1
0
0
0
Erroll Wood
Erroll Wood@errollw·
@freemocap We have nothing to share at this time unfortunately, but please stay tuned for further announcements🙂
English
0
0
1
0

Keşfet

@mc_buehler @Friggog @Microsoft @skynetislov3 @designerzen @cvtalks @afromero @Babak_Taati