high performance computing on graphics processing units: hgpu.org

Deep Learning methods usually require huge amounts of training data to perform at their full potential, and often require expensive manual labeling. Using synthetic images is therefore very attractive to train object detectors, as the labeling comes for free, and several approaches have been proposed to combine synthetic and real images for training. In this paper, we show that a simple trick is sufficient to train very effectively modern object detectors with synthetic images only. In order to train them we initialize the part responsible for feature extraction with generic layers pretrained on real data. We show that that freezing these layers and train only the remaining layers with plain OpenGL rendering performs surprisingly well.