arXiv subjects

Frédéric Champagnat

Publications and source records attributed to Frédéric Champagnat.

9 recordsLinked to original sources

arXiv2026

Getting Your Guidance Weights Right in diffusion and flow-matching posterior sampling

Training-free posterior sampling methods, also known as Plug-and-Play methods, leverage pretrained unconditional diffusion or flow-matching models to solve inverse problems. Most existing approaches rely on guidance weights to balance, at each time step, prior information from the unconditional score or velocity network with measurement consistency, yet the tuning of these weights is often not discussed and is largely left to heuristics. We introduce a simple and principled offline strategy for automatically tuning these guidance weights. Our key observation is that, at each time step, the conditional denoising score-matching objective for diffusion models, or the conditional flow-matching objective for flow-matching models, is a least-squares objective. Therefore, when the conditional prediction is expressed as a weighted sum of the unconditional network output and a measurement-guidance term, optimizing over these weights reduces to a two-dimensional linear least-squares problem. The resulting time-dependent guidance weights can be optimized offline for a given measurement operator, noise level and sampler at the cost of a single minibatch of sampling trajectories, without retraining or fine-tuning the pretrained generative model. Instantiated with the standard Tweedie-based measurement-consistency term, our approach improves posterior sampling and achieves state-of-the-art reconstruction performance across diffusion- and flow-matching-based methods. Moreover, the optimized guidance weights enable diffusion samplers to reduce the number of sampling steps from 1000 to 50 with no significant degradation in reconstruction quality. Code will be made available.

cs.LG↗

arXiv2026

Co-design approach to aperture masking for imaging through atmospheric turbulence

Aperture masking interferometry is a technique originally designed to alleviate the influence of atmospheric turbulence on images recorded on ground-based telescopes. In this communication, we explore the optimization of the aperture mask by an optical/digital co-design approach in order to obtain diffraction-limited images of relatively bright objects imaged through turbulence. We show that, with a few simplifying assumptions, it is possible to express the Mean Square Error of the restored image as a function of the chosen mask, of the spatial Power Spectral Density of the observed object and of the noise level, without actually computing any image. This allows us to optimize the aperture mask with a reduced computing cost. We also implement a multi-frame myopic algorithm to estimate jointly the observed object, the piston and the tip-tilt in front of each sub-aperture, and check by simulations that the aperture masks obtained indeed allow a satisfactory image reconstruction.

physics.optics↗

arXiv2025

Evaluating the Posterior Sampling Ability of Plug&Play Diffusion Methods in Sparse-View CT

Plug&Play (PnP) diffusion models are state-of-the-art methods in computed tomography (CT) reconstruction. Such methods usually consider applications where the sinogram contains a sufficient amount of information for the posterior distribution to be concentrated around a single mode, and consequently are evaluated using image-to-image metrics such as PSNR/SSIM. Instead, we are interested in reconstructing compressible flow images from sinograms having a small number of projections, which results in a posterior distribution no longer concentrated or even multimodal. Thus, in this paper, we aim at evaluating the approximate posterior of PnP diffusion models and introduce two posterior evaluation properties. We quantitatively evaluate three PnP diffusion methods on three different datasets for several numbers of projections. We surprisingly find that, for each method, the approximate posterior deviates from the true posterior when the number of projections decreases.

eess.IV↗

arXiv2021

Pix2Point: Learning Outdoor 3D Using Sparse Point Clouds and Optimal Transport

Good quality reconstruction and comprehension of a scene rely on 3D estimation methods. The 3D information was usually obtained from images by stereo-photogrammetry, but deep learning has recently provided us with excellent results for monocular depth estimation. Building up a sufficiently large and rich training dataset to achieve these results requires onerous processing. In this paper, we address the problem of learning outdoor 3D point cloud from monocular data using a sparse ground-truth dataset. We propose Pix2Point, a deep learning-based approach for monocular 3D point cloud prediction, able to deal with complete and challenging outdoor scenes. Our method relies on a 2D-3D hybrid neural network architecture, and a supervised end-to-end minimisation of an optimal transport divergence between point clouds. We show that, when trained on sparse point clouds, our simple promising approach achieves a better coverage of 3D outdoor scenes than efficient monocular depth methods.

cs.CV↗

arXiv2020

Translation-invariant interpolation of parametric dictionaries

In this communication, we address the problem of approximating the atoms of a parametric dictionary, commonly encountered in the context of sparse representations in "continuous" dictionaries. We focus on the case of translation-invariant dictionaries, where the inner product between atoms only depends on the difference between parameters. We investigate the following general question: is there some low-rank approximation of the dictionary $ which interpolates a subset of atoms while preserving the translation-invariant nature of the original dictionary? We derive necessary and sufficient conditions characterizing the existence of such an "interpolating" and "translation-invariant" low-rank approximation. Moreover, we provide closed-form expressions of such a dictionary when it exists. We illustrate the applicability of our results in the case of a two-dimensional isotropic Gaussian dictionary. We show that, in this particular setup, the proposed approximation framework outperforms standard Taylor approximation.

cs.IT↗

arXiv2019

Multi-Task Learning of Height and Semantics from Aerial Images

Aerial or satellite imagery is a great source for land surface analysis, which might yield land use maps or elevation models. In this investigation, we present a neural network framework for learning semantics and local height together. We show how this joint multi-task learning benefits to each task on the large dataset of the 2018 Data Fusion Contest. Moreover, our framework also yields an uncertainty map which allows assessing the prediction of the model. Code is available at https://github.com/marcelampc/mtl_aerial_images .

cs.CV↗

arXiv2019

Double frame Tomographic PTV at high seeding densities

A novel method performing 3D PTV from double frame multi-camera images is introduced. Particle velocities are estimated by following three steps. Firstly, separate particle reconstructions with a sparsity-based algorithm are performed on a fine grid. Secondly, they are expanded on a coarser grid on which 3D correlation is performed, yielding a predictor displacement field that allows to efficiently match particles at the two time instants. As these particles are still located on a voxel grid, the third, final step achieves particle position refinement to their actual subvoxel position by a global optimization process, also accounting for their intensities. As it strongly leverages on principles from tomographic reconstruction, the technique is termed Double-Frame Tomo-PTV (DF-TPTV). Synthetic tests on a complex turbulent flow show that the method achieves high particle and vector detection efficiency, up to seeding densities of around 0.08 particles per pixel (ppp), while its root-mean-square error of velocity estimation is lower to that of state-of-the-art similar methods. Results from an experimental campaign on a transitional round air jet at Reynolds number 4600 are also presented. During the tests, seeding density varies from 0.06 to 0.03 ppp on average. Associated to an outlier rejection scheme based on temporal statistics, DF-TPTV vector fields truthfully correspond to the instantaneous jet dynamics. Quantitative performance assessment is provided by introducing statistics performed by bin averaging, upon assuming statistical axisymmetry of the jet. Mean and fluctuating axial velocity components in the jet near-field are compared with reference results obtained from planar PIV at higher seeding density.

physics.flu-dyn↗

arXiv2018

Deep Depth from Defocus: how can defocus blur improve 3D estimation using dense neural networks?

Depth estimation is of critical interest for scene understanding and accurate 3D reconstruction. Most recent approaches in depth estimation with deep learning exploit geometrical structures of standard sharp images to predict corresponding depth maps. However, cameras can also produce images with defocus blur depending on the depth of the objects and camera settings. Hence, these features may represent an important hint for learning to predict depth. In this paper, we propose a full system for single-image depth prediction in the wild using depth-from-defocus and neural networks. We carry out thorough experiments to test deep convolutional networks on real and simulated defocused images using a realistic model of blur variation with respect to depth. We also investigate the influence of blur on depth prediction observing model uncertainty with a Bayesian neural network approach. From these studies, we show that out-of-focus blur greatly improves the depth-prediction network performances. Furthermore, we transfer the ability learned on a synthetic, indoor dataset to real, indoor and outdoor images. For this purpose, we present a new dataset containing real all-focus and defocused images from a Digital Single-Lens Reflex (DSLR) camera, paired with ground truth depth maps obtained with an active 3D sensor for indoor scenes. The proposed approach is successfully validated on both this new dataset and standard ones as NYUv2 or Depth-in-the-Wild. Code and new datasets are available at https://github.com/marcelampc/d3net_depth_estimation

cs.CV↗

arXiv2009

Point target detection and subpixel position estimation in optical imagery

This paper addresses the issue of detecting point objects in a clutter background and estimating their position by image processing. We are interested in the specific context where the object signature significantly varies with its random subpixel location because of aliasing. Conventional matched filter neglects this phenomenon and causes consistent loss of detection performance. Thus, alternative detectors are proposed and numerical results show the improvement brought by approximate and generalized likelihood ratio tests in comparison with pixel matched filtering. We also study the performance of two types of subpixel position estimators. Finally, we put forward the major influence of sensor design on both estimation and point object detection performance.

stat.AP↗