Using AI for Real-Time 3D Face Reconstruction
Events
Subscribe:  iCal  |  Google Calendar
Amsterdam NL   25, Jun — 28, Jun
Los Angeles US   25, Jun — 28, Jun
Montreal CA   27, Jun — 1, Jul
Cambridge GB   28, Jun — 2, Jul
Guildford GB   29, Jun — 30, Jun
Latest comments
by Matthew Scenery.Melbourne
2 hours ago

Their website does say that you can pay per image at $1 per image. I am in the opposite boat though. I could see this having a very significant effect on photogrammetry but I would need to process a few thousand images at a time which would not be very feasible with their current pricing model

by Shaun
3 hours ago

OMFG....PLEEEEEEEEEEEEEEEASE!

To the developers. A very promising piece of software for a VFX supervisor like me. BUT, please reconsider your pricing tiers and introduce a per-image price. We are a pretty large facility, but I can only imagine needing about 1-10 images a month at the very most. It's like HDRI's - we buy them all the time, one at a time. They need to be individually billed so a producer can charge them against a particular job.

Using AI for Real-Time 3D Face Reconstruction
27 April, 2018
News

What is your take on facial alignment and real-time 3D face reconstruction? There are so many possibilities here, but are such things even possible today? Take a look at a work that proposes using a convolutional neural network to deal with the tasks. The method is said to be capable of generating high-quality outputs, creating each image in less than 10 milliseconds. What’s the magic here?

First, let’s check out a quick introduction from Two Minute Papers:

Here is an abstract: 

Joint 3D Face Reconstruction and Dense Alignment with Position Map Regression Network 

We propose a straightforward method that simultaneously reconstructs the 3D facial structure and provides dense alignment. To achieve this, we design a 2D representation called UV position map which records the 3D shape of a complete face in UV space, then train a simple Convolutional Neural Network to regress it from a single 2D image. We also integrate a weight mask into the loss function during training to improve the performance of the network. Our method does not rely on any prior face model and can reconstruct full facial geometry along with semantic meaning. Meanwhile, our network is very light-weighted and spends only 9.8ms to process an image, which is extremely faster than previous works. Experiments on multiple challenging datasets show that our method surpasses other state-of-the-art methods on both reconstruction and alignment tasks by a large margin.

Yao FengFan WuXiaohu ShaoYanfeng WangXi Zhou 

You can find the full paper here

Links: 

What are your thoughts on the proposed method? What do you think about its limitations? 

Source: arxiv.org

Leave a Reply

avatar