A data-driven compression method for transient rendering

Abstract

Monte Carlo methods for transient rendering have become a powerful instrument to generate reliable data in transient imaging applications, either for benchmarking, analysis, or as a source for data-driven approaches. However, due to the increased dimensionality of time-resolved renders, storage and data bandwidth are signi cant limiting constraints, where a single time-resolved render of a scene can take several hundreds of megabytes. In this work we propose a learning-based approach that makes use of deep encoderdecoder architectures to learn lower-dimensional feature vectors of time-resolved pixels. We demonstrate how our method is capable of compressing transient renders up to a factor of 32, and recover the full transient pro le making use of a decoder. Additionally, we show how our learned features signi cantly mitigate variance on the recovered signal, addressing one of the pathological problems in transient rendering. CCS CONCEPTS •Computing methodologies → Ray tracing;

Efficient Large-Scale GPS Trajectory Compression on Spark: A Pipeline-Based Approach

“…T and the timestamp when compression is finished is 2 T , then the compression time T can be defined as Equation ( 7):…”

“…Second, the massive GPS trajectory data requires many computing resources when the owners perform data analysis and mining tasks. Third, the massive amount of data renders traditional visualization methods ineffective [2]. In order to alleviate the above pressure, GPS trajectories need to be compressed.…”

“…This allows the operator to finely customize and control the data-processing workflow according to specific requirements. (2) We parallelize a set of classical algorithms in Spark to meet the requirements of compressing large-scale GPS trajectory dataset. These algorithms consist of the DP, TD-TR, SW, SQUISH, and V-DP.…”

Lightweight Deep Learning Architecture for MPI Correction and Transient Reconstruction

“…They then proposed an algorithm for transient reconstruction from a high number of iToF modulation frequencies. On a different note, Liang et al [29] devised a deep learning model for the compression of rendered transient data, an important task due to the high volume of the data and the dangerous amount of rendering noise.…”

“…The complexity of the matter makes an encoding of the ground truth a necessity. Since the one proposed in[15] is way too simplistic, and the one by Liang et al[31] computationally heavy, we propose a novel approximation of the transient vectorx g with just 6 parameters, 2 needed for the direct component, and the other 4 for the global. Therefore, the Transient Reconstruction Module is further split into two components, the Direct Model that takes care of the reconstruction of the direct component and the Global Model, which instead predicts the global component.…”