Synthetic for Computer Vision

This is a repo for tracking the progress of using synthetic images for computer vision research. If you found any important work is missing or information is not up-to-date, please edit this file directly and make a pull request. Each publication is tagged with a keyword to make it easier to search.

If you find anything missing from this page, please edit this README.md file to add it. When adding a new item, you can simply follow the format of existing items. How this document is structured is documented in contribute.md.

How to use: Click publication to jump to the paper title, detailed information such as code and project page will be provided together with pdf file.**

Synthetic image dataset

• SunCG (Princeton)
• Minos
• House3d (Facebook)
• Procedural Human Action Videos (PHAV)
• SURREAL
• Virtual KITTI
• Synthetia
• Sintel, A synthetic dataset for optical flow
• SceneFlow
• 4D Light Fields
• ICL-NUIM dataset
• Driving in the Matrix

3D Model Repository

Realistic 3D models are critical for creating realistic and diverse virtual worlds. Here are research efforts for creating 3D model repositories.

• ShapeNet
• 3dscan
• seeing3Dchairs

Tools

• AirSim (Microsoft)
• CARLA (Intel)
• Unity ML agents
• Render SMPL human bodies on Blender, see CVPR2017
• Render for CNN, based on Blender, see ICCV2015
• UETorch, based on UE4, see ICML2016
• UnrealCV, based on UE4, see ArXiv
• VizDoom, based on Doom, see ArXiv
• OpenAI Universe, see project page
• Blender addon for 4D light field rendering, see project page

Resources

ECCV 2016 Workshop Virtual/Augmented Reality for Visual Artificial Intelligence (VARVAI) workshop

ICCV 2017 Workshop Role of Simulation in Computer Vision

Virtual Reality Meets Physical Reality: Modelling and Simulating Virtual Humans and Environments Siggraph Asia 2016 workshop

CVPR 2017 Workshop THOR Challenge

See also: http://riemenschneider.hayko.at/vision/dataset/index.php?filter=+synthetic

Misc.

• RealismCNN github
• Abnormality Detection in Images(http://paul.rutgers.edu/~babaks/abnormality_detection.html)

Reference

2017

(Total=11)

• Adversarially Tuned Scene Generation (pdf)

• UE4Sim: A Photo-Realistic Simulator for Computer Vision Applications (pdf) (project)

• Playing for Benchmarks (pdf)

• A Self-supervised Learning System for Object Detection using Physics Simulation and Multi-view Pose Estimation (code) (pdf) (project)

• Procedural Generation of Videos to Train Deep Action Recognition Networks (pdf) (project) (citation:8)

• Learning from Synthetic Humans (code) (pdf) (project) tag: synthetic human

• Nvidia Issac

• Configurable, Photorealistic Image Rendering and Ground Truth Synthesis by Sampling Stochastic Grammars Representing Indoor Scenes

• Aerial Informatics and Robotics Platform (code) (pdf) (project) tag: tool

• Tobin, Josh, et al. "Domain Randomization for Transferring Deep Neural Networks from Simulation to the Real World." arXiv preprint arXiv:1703.06907 (2017). tag: domain (pdf)

• M. Johnson-Roberson, C. Barto, R. Mehta, S. N. Sridhar, Karl Rosaen,and R. Vasudevan, “Driving in the matrix: Can virtual worlds replace human-generated annotations for real world tasks?,” in IEEE International Conference on Robotics and Automation, pp. 1–8, 2017. (code) (pdf) (project) (citation:3)

2016

(Total=17)

• Sadeghi, Fereshteh, and Sergey Levine. "rl: Real single-image flight without a single real image. arXiv preprint." arXiv preprint arXiv:1611.04201 12 (2016). tag: rl

• Johnson, Justin, et al. "CLEVR: A Diagnostic Dataset for Compositional Language and Elementary Visual Reasoning." arXiv preprint arXiv:1612.06890 (2016). (pdf)

• McCormac, John, et al. "SceneNet RGB-D: 5M Photorealistic Images of Synthetic Indoor Trajectories with Ground Truth." arXiv preprint arXiv:1612.05079 (2016).

• de Souza, César Roberto, et al. "Procedural Generation of Videos to Train Deep Action Recognition Networks." arXiv preprint arXiv:1612.00881 (2016). (pdf) (project) tag: synthetic human

• Synnaeve, Gabriel, et al. "TorchCraft: a Library for Machine Learning Research on Real-Time Strategy Games." arXiv preprint arXiv:1611.00625 (2016). (pdf) (code)

• Lin, Jenny, et al. "A virtual reality platform for dynamic human-scene interaction." SIGGRAPH ASIA 2016 Virtual Reality meets Physical Reality: Modelling and Simulating Virtual Humans and Environments. ACM, 2016. (pdf) (project)

• Mahendran, A., et al. "ResearchDoom and CocoDoom: Learning Computer Vision with Games." arXiv preprint arXiv:1610.02431 (2016). (pdf) (project)

• The SYNTHIA dataset: A large collection of synthetic images for semantic segmentation of urban scenes. 2016 (pdf) (project) (citation:4)

• Virtual Worlds as Proxy for Multi-Object Tracking Analysis. 2016
  (pdf) (project) (citation:5)

• Playing for data: Ground truth from computer games. 2016
  (pdf) (citation:1)

• Play and Learn: Using Video Games to Train Computer Vision Models. 2016
  (pdf) (citation:1)

• ViZDoom: A Doom-based AI Research Platform for Visual Reinforcement Learning. 2016
  (code) (pdf) (project) (citation:4)

• A large dataset of object scans. 2016
  (pdf) (project) (citation:6)

• UnrealCV: Connecting Computer Vision to Unreal Engine 2016
  (code) (project) (pdf)

• Learning Physical Intuition of Block Towers by Example 2016
  (code) (pdf) (citation:12)

• Target-driven Visual Navigation in Indoor Scenes using Deep Reinforcement Learning 2016
  (pdf)

• A Dataset and Evaluation Methodology for Depth Estimation on 4D Light Fields. ACCV 2016
  (code) (pdf) (project) (citation)

2015

(Total=3)

• A Large Dataset to Train Convolutional Networks for Disparity, Optical Flow, and Scene Flow Estimation. 2015
  (pdf) (citation:9)

• Render for cnn: Viewpoint estimation in images using cnns trained with rendered 3d model views. 2015
  (code) (pdf) (citation:33)

• Shapenet: An information-rich 3d model repository. 2015
  (pdf) (project) (citation:27)

2014

(Total=2)

• Virtual and real world adaptation for pedestrian detection. 2014
  (pdf) (citation:46)

• Seeing 3d chairs: exemplar part-based 2d-3d alignment using a large dataset of cad models. 2014
  (code) (pdf) (project) (citation:110)

• Handa, Ankur, Thomas Whelan, John McDonald, and Andrew J. Davison. "A benchmark for RGB-D visual odometry, 3D reconstruction and SLAM." In Robotics and automation (ICRA), 2014 IEEE international conference on, pp. 1524-1531. IEEE, 2014. (project)

2013

(Total=1)

• Detailed 3d representations for object recognition and modeling. 2013
  (pdf) (citation:67)

2012

(Total=1)

• A naturalistic open source movie for optical flow evaluation. 2012
  (pdf) (project) (citation:227)

2010

(Total=1)

• Learning appearance in virtual scenarios for pedestrian detection. 2010
  (pdf) (citation:79)

2007

(Total=1)

• Ovvv: Using virtual worlds to design and evaluate surveillance systems. 2007
  (pdf) (citation:58)