Learning to Generate Synthetic 3D Training Data through Hybrid Gradient
This work addresses the problem of efficiently generating effective synthetic training data for computer vision tasks like surface normal estimation, which is incremental as it builds on existing optimization methods.
The paper tackles the challenge of generating synthetic 3D training data that improves deep network performance on real images by proposing a hybrid gradient method to optimize design decisions like 3D shape selection and camera placement. It shows that this approach outperforms prior state-of-the-art methods in optimizing 3D training data generation, with notable gains in computational efficiency.
Synthetic images rendered by graphics engines are a promising source for training deep networks. However, it is challenging to ensure that they can help train a network to perform well on real images, because a graphics-based generation pipeline requires numerous design decisions such as the selection of 3D shapes and the placement of the camera. In this work, we propose a new method that optimizes the generation of 3D training data based on what we call "hybrid gradient". We parametrize the design decisions as a real vector, and combine the approximate gradient and the analytical gradient to obtain the hybrid gradient of the network performance with respect to this vector. We evaluate our approach on the task of estimating surface normal, depth or intrinsic decomposition from a single image. Experiments on standard benchmarks show that our approach can outperform the prior state of the art on optimizing the generation of 3D training data, particularly in terms of computational efficiency.