12 megapixel 16-bit linear input (tone-mapped for visualization) tone-mapped with HDR+ 400 -600 ms processed with our algorithm 61 ms, PSNR = 28.4 dB

FlexISP 32.5 dB Ours 38.4 dB Adobe CR 31.7 dB reference noisy ours 33.3 dB ref. [Condat 2012] 32.4 dB Figure 1:We propose a data-driven approach for jointly solving denoising and demosaicking. By carefully designing a dataset made of rare but challenging image features, we train a neural network that outperforms both the state-of-the-art and commercial solutions on demosaicking alone (group of images on the left, insets show error maps), and on joint denoising-demosaicking (on the right, insets show close-ups). The benefit of our method is most noticeable on difficult image structures that lead to moiré or zippering of the edges.

ConvNetQuake is the first neural network for detection and location of earthquakes from seismograms.

We present a new algorithm to automatically schedule Halide programs for high-performance image processing and deep learning. We significantly improve upon the performance of previous methods, which considered a limited subset of schedules. We define a parameterization of possible schedules much larger than prior methods and use a variant of beam search to search over it. The search optimizes runtime predicted by a cost model based on a combination of new derived features and machine learning. We train the cost model by generating and featurizing hundreds of thousands of random programs and schedules. We show that this approach operates effectively with or without autotuning. It produces schedules which are on average almost twice as fast as the existing Halide autoscheduler without autotuning, or more than twice as fast with, and is the first automatic scheduling algorithm to significantly outperform human experts on average. CCS Concepts: • Computing methodologies → Image processing; • Software and its engineering → Domain specific languages.

and Inria (a) our algorithm can relight a single-illumination drone video dynamically to synthesize a "time-lapse" e ect (b) single-view input (c) three relit outputs: here we built the proxy geometry using internet photos of the same location