Automatic 2D-to-3D Video Conversion with CNNs

Related tags

Deep Learningdeep3d
Overview

Deep3D: Automatic 2D-to-3D Video Conversion with CNNs

How To Run

To run this code. Please install MXNet following the official document. Deep3D requires MXNet to be built with Cuda 7.0 and Cudnn 4 or above. Please open mxnet/config.mk and set USE_CUDA and USE_CUDNN to 1. Then, append EXTRA_OPERATORS=path/to/deep3d/operators to path/to/mxnet/config.mk and recompile MXNet.

alt text

Motivation

Since the debut of Avatar in 2008, 3D movies has rapidly developed into mainstream technology. Roughly 10 to 20 3D movies are produced each year and the launch of Oculus Rift and other VR head set is only going to drive up the demand.

Producing 3D movies, however, is still hard. There are two ways of doing this and in practice they are about equally popular: shooting with a special 3D camera or shooting in 2D and manually convert to 3D. But 3D cameras are expensive and unwieldy while manual conversion involves an army of "depth artists" who sit there and draw depth maps for each frame.

Wouldn't it be cool if 2D-to-3D conversion can be done automatically, if you can take a 3D selfie with an ordinary phone?

Teaser

In case you are already getting sleepy, here are some cool 3D images converted from 2D ones by Deep3D. Normally you need 3D glasses or VR display to watch 3D images, but since most readers won't have these we show the 3D images as GIFs.

alt text alt text alt text alt text alt text alt text alt text alt text

Method

3D imagery has two views, one for the left eye and the other for the right. To convert an 2D image to 3D, you need to first estimate the distance from camera for each pixel (a.k.a depth map) and then wrap the image based on its depth map to create two views.

The difficult step is estimating the depth map. For automatic conversion, we would like to learn a model for it. There are several works on depth estimation from single 2D image with DNNs. However, they need to be trained on image-depth pairs which are hard to collect. As a result they can only use small datasets with a few hundred examples like NYU Depth and KITTI. Moreover, these datasets only has static scenes and it's hard to imagine they will generalize to photos with people in them.

In Contrast, Deep3D can be trained directly on 3D movies that have tens of millions frames in total. We do this by making the depth map an internal representation instead of the end prediction. Thus, instead of predicting an depth map and then use it to recreate the missing view with a separate algorithm, we train depth estimation and recreate end-to-end in the same neural network.

Here are some visualizations of our internal depth representation to help you understand how it works:

alt text alt text alt text alt text alt text alt text alt text alt text alt text

Following each image, there are 4-by-3 maps of depth layers, ordered from near to far. You can see that objects that are near to you appear in the first depth maps and objects that are far away appear in the last ones. This shows that the internal depth representation is learning to infer depth from 2D images without been directly trained on it.

Code

This work is done with MXNet, a flexible and efficient deep learning package. The trained model and a prediction script is in deep3d.ipynb. We will release the code for training shortly.

Owner
Eric Junyuan Xie
Software Engineer @ Bytedance
Eric Junyuan Xie
Posterior temperature optimized Bayesian models for inverse problems in medical imaging

Posterior temperature optimized Bayesian models for inverse problems in medical imaging Max-Heinrich Laves*, Malte Tölle*, Alexander Schlaefer, Sandy

Artificial Intelligence in Cardiovascular Medicine (AICM) 6 Sep 19, 2022
Reimplementation of the paper `Human Attention Maps for Text Classification: Do Humans and Neural Networks Focus on the Same Words? (ACL2020)`

Human Attention for Text Classification Re-implementation of the paper Human Attention Maps for Text Classification: Do Humans and Neural Networks Foc

Shunsuke KITADA 15 Dec 13, 2021
Official code for 'Pixel-wise Energy-biased Abstention Learning for Anomaly Segmentationon Complex Urban Driving Scenes'

PEBAL This repo contains the Pytorch implementation of our paper: Pixel-wise Energy-biased Abstention Learning for Anomaly Segmentation on Complex Urb

Yu Tian 117 Jan 03, 2023
A Survey on Deep Learning Technique for Video Segmentation

A Survey on Deep Learning Technique for Video Segmentation A Survey on Deep Learning Technique for Video Segmentation Wenguan Wang, Tianfei Zhou, Fati

Tianfei Zhou 112 Dec 12, 2022
Sarus implementation of classical ML models. The models are implemented using the Keras API of tensorflow 2. Vizualization are implemented and can be seen in tensorboard.

Sarus published models Sarus implementation of classical ML models. The models are implemented using the Keras API of tensorflow 2. Vizualization are

Sarus Technologies 39 Aug 19, 2022
SBINN: Systems-biology informed neural network

SBINN: Systems-biology informed neural network The source code for the paper M. Daneker, Z. Zhang, G. E. Karniadakis, & L. Lu. Systems biology: Identi

Lu Group 15 Nov 19, 2022
Implementation of the paper titled "Using Sampling to Estimate and Improve Performance of Automated Scoring Systems with Guarantees"

Using Sampling to Estimate and Improve Performance of Automated Scoring Systems with Guarantees Implementation of the paper titled "Using Sampling to

MIDAS, IIIT Delhi 2 Aug 29, 2022
Python code to fuse multiple RGB-D images into a TSDF voxel volume.

Volumetric TSDF Fusion of RGB-D Images in Python This is a lightweight python script that fuses multiple registered color and depth images into a proj

Andy Zeng 845 Jan 03, 2023
A simple, fast, and efficient object detector without FPN

You Only Look One-level Feature (YOLOF), CVPR2021 A simple, fast, and efficient object detector without FPN. This repo provides an implementation for

789 Jan 09, 2023
BlockUnexpectedPackets - Preventing BungeeCord CPU overload due to Layer 7 DDoS attacks by scanning BungeeCord's logs

BlockUnexpectedPackets This script automatically blocks DDoS attacks that are sp

SparklyPower 3 Mar 31, 2022
TeachMyAgent is a testbed platform for Automatic Curriculum Learning methods in Deep RL.

TeachMyAgent: a Benchmark for Automatic Curriculum Learning in Deep RL Paper Website Documentation TeachMyAgent is a testbed platform for Automatic Cu

Flowers Team 51 Dec 25, 2022
PyTorch code for our paper "Gated Multiple Feedback Network for Image Super-Resolution" (BMVC2019)

Gated Multiple Feedback Network for Image Super-Resolution This repository contains the PyTorch implementation for the proposed GMFN [arXiv]. The fram

Qilei Li 66 Nov 03, 2022
PyTorch3D is FAIR's library of reusable components for deep learning with 3D data

Introduction PyTorch3D provides efficient, reusable components for 3D Computer Vision research with PyTorch. Key features include: Data structure for

Facebook Research 6.8k Jan 01, 2023
The Video-based Accident Detection System built in Python

Accident-detection-system About the Project This Repository contains the Video-based Accident Detection System built in Python. Contributors Yukta Gop

SURYAVANSHI SNEHAL BALKRISHNA 50 Dec 07, 2022
Source code for our paper "Learning to Break Deep Perceptual Hashing: The Use Case NeuralHash"

Learning to Break Deep Perceptual Hashing: The Use Case NeuralHash Abstract: Apple recently revealed its deep perceptual hashing system NeuralHash to

<a href=[email protected]"> 11 Dec 03, 2022
Consecutive-Subsequence - Simple software to calculate susequence with highest sum

Simple software to calculate susequence with highest sum This repository contain

Gbadamosi Farouk 1 Jan 31, 2022
A Momentumized, Adaptive, Dual Averaged Gradient Method for Stochastic Optimization

MADGRAD Optimization Method A Momentumized, Adaptive, Dual Averaged Gradient Method for Stochastic Optimization pip install madgrad Try it out! A best

Meta Research 774 Dec 31, 2022
Pytorch reimplementation of PSM-Net: "Pyramid Stereo Matching Network"

This is a Pytorch Lightning version PSMNet which is based on JiaRenChang/PSMNet. use python main.py to start training. PSM-Net Pytorch reimplementatio

XIAOTIAN LIU 1 Nov 25, 2021
Edge-aware Guidance Fusion Network for RGB-Thermal Scene Parsing

EGFNet Edge-aware Guidance Fusion Network for RGB-Thermal Scene Parsing Dataset and Results Test maps: 百度网盘 提取码:zust Citation @ARTICLE{ author={Zhou,

ShaohuaDong 10 Dec 08, 2022
Interacting Two-Hand 3D Pose and Shape Reconstruction from Single Color Image (ICCV 2021)

Interacting Two-Hand 3D Pose and Shape Reconstruction from Single Color Image Interacting Two-Hand 3D Pose and Shape Reconstruction from Single Color

75 Dec 02, 2022