RealSynth

Photorealistic Synthetic Image Data and Ground Truth

Ground Truth
Synthesized

Synthetic eye candy visuals please not only humans but also machine learning processes. Our fully automated pixel perfect segmentation capturing process is optimized for maximum training performance.

A variety of industry standard camera and lens presets can be selected. Any motion of the camera carrier is freely customizable and deterministic to provide reproducible scenario captures with varying setups. The capturing process outputs thousands of fully annotated images within a few minutes.

Fully Custom Scenarios

Robust classifiers require a large variety of inputs during training. To achieve maximum performance in every real-life situation, labeled data at different times of day and various weather conditions are needed. Each scenario can be equipped with the ability to change these environmental settings in an instant.

Support for Various Annotation Standards

Labeled output can be easily customized to various formats. Currently we support labeled output that is directly compatible with quasi-standard industry machine learning data formats like KITTI, MS COCO and Pascal VOC. Therefore, the data can seamlessly be fed into already established workflows and commonly used machine learning tools and libraries.

Methodology

Training on synthetic data, rather than on actually captured and annotated real images is an emerging trend in recent computer vision research. Studies have shown than classifiers and detectors that were trained exclusively on synthetic data can perform almost as good as those that were trained on large manually labeled data sets [1,3]. This alone is a remarkable achievement, considering how expensive real data is compared to synthetic data. Additionally, if both synthetic and real data is combined during training, a gain in performance can be expected [2].

  1. Biliana Kaneva, Antonio Torralba, William T. Freeman. (2011) Evaluating Image Feaures Using a Photorealistic Virtual World. In Proceedings of the 2011 IEEE International Conference on Computer Vision (ICCV)
  2. German Ros, Laura Sellart, Joanna Materzynska, David Vazquez, Antonio Lopez. (2016) The SYNTHIA Dataset: A Large Collection of Synthetic Images for Semantic Segmentation of Urban Scenes. In Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)
  3. Stephan R. Richter, Vibhav Vineet, Stefan Roth, Vladlen Koltun. (2016) Playing for Data: Ground Truth from Computer Games. In Leibe B., Matas J., Sebe N., Welling M. (eds) Computer Vision – ECCV 2016. ECCV 2016. Lecture Notes in Computer Science, vol 9906. Springer, Cham