The Fast Bilateral Solver

By Jonathan T. Barron, Ben Poole. European Conference on Computer Vision, 2015

403 citations recorded by the source corpus.

We present the bilateral solver, a novel algorithm for edge-aware smoothing that combines the flexibility and speed of simple filtering approaches with the accuracy of domain-specific optimization algorithms. Our technique is capable of matching or improving upon state-of-the-art results on several different computer vision tasks (stereo, depth superresolution, colorization, and semantic segmentation) while being 10–1000\(\times \) faster than baseline techniques with comparable accuracy, and producing lower-error output than techniques with comparable runtimes. The bilateral solver is fast, robust, straightforward to generalize to new domains, and simple to integrate into deep learning pipelines.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/6a16d519a5ce9a27cefd11c4f96c714fd2413c50

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Inverse Rendering for Complex Indoor Scenes: Shape, Spatially-Varying Lighting and SVBRDF From a Single Image (2019, methodology): “Refinement using bilateral solver We use an end-to-end trainable bilateral solver to impose a smoothness prior [5, 31].”
Non-Local Spatial Propagation Network for Depth Completion (2020, neutral): “ we provide results from the Sparse-to-Dense Non-Local Spatial Propagation Network for Depth Completion 11 Method RMSE (m) REL 1:25 1:252 1:253 S2D [21] 0.230 0.044 97.1 99.4 99.8 [21]+Bilateral [4] 0.479 0.084 92.4 97.6 98.9 [21]+SPN [19] 0.172 0.031 98.3 99.7 99.9 DepthCoeff [14] 0.118 0.013 99.”
Self-Supervised Sparse-to-Dense: Self-Supervised Depth Completion from LiDAR and Monocular Camera (2018, background): “, Microsoft Kinect) [11] is sometimes also referred to as depth inpainting [12], or depth enhancement [5–7] when noise is taken into account.”
PixelNet: Towards a General Pixel-level Architecture (2016, background): “…pixel, numerous approaches have explored post-processing steps that enforce spatial consistency across labels via e.g., bilateral smoothing with fully-connected Gaussian CRFs [13, 40, 74] or bilateral solvers [5], dilated spatial convolutions [73], LSTMs [10], and convolutional pseudo priors [70].”
Learning Sparse High Dimensional Filters: Image Filtering, Dense CRFs and Bilateral Neural Networks (2015, background): “The bilateral filter also has been proposed to regularize a large set of applications in [9, 8] and the respective optimization problems are parametrized in a bilateral space.”
Depth-aware CNN for RGB-D Segmentation (2018, methodology): “Bilateral filters [3,2] is widely used in computer graphics for image smoothness with edge preserving.”
Deep Depth Completion of a Single RGB-D Image (2018, methodology): “As such, the methods we consider include: a) joint bilinear ﬁltering [64] (Bilateral), b) fast bilateral solver [5] (Fast), and c) global edge-aware energy optimization [20] (TGV).”
Simple Does It: Weakly Supervised Instance and Semantic Segmentation (2016, background): “This inspires us to follow a recursive training procedure, where these new labels are fed in as ground truth for a second training round.”
Semantic Segmentation With Context Encoding and Multi-Path Decoding (2020, background): “Apart from CRF, Barron and Poole [4] proposed bilateral solver to prompt edge-aware smoothness, and Jampani et al.”
Normalized Cut Loss for Weakly-Supervised CNN Segmentation (2018, background): “Another interesting direction is to explore MRF/CRF regularizer based losses and their different relaxations [28, 3, 46] discussed in our follow-up work [47].”
One-Shot Video Object Segmentation (2016, methodology): “First, we propose the use of the Fast Bilateral Solver (FBS) [2] to snap the background prediction to the image edges.”
NICE-SLAM: Neural Implicit Scalable Encoding for SLAM (2021, methodology): “For fair comparison, we apply the bilateral solver [2] to DI-Fusion [16] and TSDFFusion to fill depth holes before calculating the average L1 loss.”
Learning to reconstruct shape and spatially-varying reflectance from a single image (2018, background): “Such a prior might be explicitly enhanced to improve performance by using a densely connected CRF [Ristovski et al. 2013] or bilateral filter [Barron and Poole 2016].”
Learning Depth with Convolutional Spatial Propagation Network (2018, methodology): “using bilateral filtering [24] as a post-processing module”
Deepstrip: High-Resolution Boundary Refinement (2020, methodology): “For upsampling based approaches, we compare our approach with Bilinear Upsampling , Bilateral Solver [6] , Joint Bilateral Upsampling [20] ( JBU ), Guided Filtering [17] and Deep GF [40] .”
Real-time user-guided image colorization with learned deep priors (2017, methodology): “Baselines (Barron and Poole 2016; Endo et al. 2016; Levin et al. 2004) are computed with publicly available code from the authors.”
Fast End-to-End Trainable Guided Filter (2018, methodology): “al ﬁlter usually requires a large amount of computation resources. Thus, many methods [28]– [30] are presented to reduce the computation complexity. Build on joint bilateral upsampling, Barron et al. [31] present a new form of bilateral-space optimization that efﬁciently solves a regularized least-s”
Dense and Low-Rank Gaussian CRFs Using Deep Embeddings (2017, methodology): “Regarding (c), [2] showed that efficient exact inference can be used for the sparselyconnected case using conjugate gradient, while [1] showed that for densely-connected graphs with bilateral-type pairwise terms linear system methods can be used for efficient inference and backpropagation.”