No abstract

We present a new deep supervised learning method for intrinsic decomposition of a single image into its albedo and shading components. Our contributions are based on a new fully convolutional neural network that estimates absolute albedo and shading jointly. Our solution relies on a single end-to-end deep sequence of residual blocks and a perceptually-motivated metric formed by two adversarially trained discriminators.As opposed to classical intrinsic image decomposition work, it is fully data-driven, hence does not require any physical priors like shading smoothness or albedo sparsity, nor does it rely on geometric information such as depth. Compared to recent deep learning techniques, we simplify the architecture, making it easier to build and train, and constrain it to generate a valid and reversible decomposition. We rediscuss and augment the set of quantitative metrics so as to account for the more challenging recovery of non scale-invariant quantities.We train and demonstrate our architecture on the publicly available MPI Sintel dataset and its intrinsic image decomposition, show attenuated overfitting issues and discuss generalizability to other data. Results show that our work outperforms the state of the art deep algorithms both on the qualitative and quantitative aspect.

Figure 1: An end-to-end convolutional neural network (CNN) trained without ground truth supervision decomposes an input image into its albedo and shading images. In a pre-computation step (top), a unique CNN is trained by independently processing two images (taken from a time-varying sequence) so that a loss function can be expressed in a siamese manner on both their decompositions comparatively (i.e., red background): it can learn from observing the changes in the input's shading. At runtime (bottom), the trained CNN offers decomposition of previously unseen single images.As a result, our method has the good properties of i) arXiv:1803.00805v2 [cs.CV] 3 Sep 2018• Additionally, in section 4, we propose a synthetic dataset that is generated using physically-based rendering. It is an extension of the recent SUNCG dataset [29], which we augment for unsupervised SIID by rendering static scenes under varying lighting conditions and tone mappings.

We propose a new approach for detecting repeated patterns on a grid in a single image. To do so, we detect repetitions in the space of pre-trained deep CNN filter responses at all layer levels. These encode features at several conceptual levels (from low-level patches to high-level semantics) as well as scales (from local to global). As a result, our repeated pattern detector is robust to challenging cases where repeated tiles show strong variation in visual appearance due to occlusions, lighting or background clutter. Our method contrasts with previous approaches that rely on keypoint extraction, description and clustering or on patch correlation. These generally only detect low-level feature clusters that do not handle variations in visual appearance of the patterns very well. Our method is simpler, yet incorporates high level features implicitly. As such, we can demonstrate detections of repetitions with strong appearance variations, organized on a nearly-regular axis-aligned grid Results show robustness and consistency throughout a varied database of more than 150 images.

No abstract