Search

While the field of inverse graphics has been witnessing continuous growth, techniques devised thus far predominantly focus on learning individual scene representations. In contrast, learning large sets of scenes has been a considerable bottleneck in NeRF developments, as repeatedly applying inverse graphics on a sequence of scenes, though essential for various applications, remains largely prohibitive in terms of resource costs. We introduce a framework termed "scaled inverse graphics", aimed at efficiently learning large sets of scene representations, and propose a novel method to this end. It operates in two stages: (i) training a compression model on a subset of scenes, then (ii) training NeRF models on the resulting smaller representations, thereby reducing the optimization space per new scene. In practice, we compact the representation of scenes by learning NeRFs in a latent space to reduce the image resolution, and sharing information across scenes to reduce NeRF representation complexity. We experimentally show that our method presents both the lowest training time and memory footprint in scaled inverse graphics compared to other methods applied independently on each scene. Our codebase is publicly available as open-source. Our project page can be found at https://scaled-ig.github.io .

While pre-trained image autoencoders are increasingly utilized in computer vision, the application of inverse graphics in 2D latent spaces has been under-explored. Yet, besides reducing the training and rendering complexity, applying inverse graphics in the latent space enables a valuable interoperability with other latent-based 2D methods. The major challenge is that inverse graphics cannot be directly applied to such image latent spaces because they lack an underlying 3D geometry. In this paper, we propose an Inverse Graphics Autoencoder (IG-AE) that specifically addresses this issue. To this end, we regularize an image autoencoder with 3D-geometry by aligning its latent space with jointly trained latent 3D scenes. We utilize the trained IG-AE to bring NeRFs to the latent space with a latent NeRF training pipeline, which we implement in an open-source extension of the Nerfstudio framework, thereby unlocking latent scene learning for its supported methods. We experimentally confirm that Latent NeRFs trained with IG-AE present an improved quality compared to a standard autoencoder, all while exhibiting training and rendering accelerations with respect to NeRFs trained in the image space. Our project page can be found at https://ig-ae.github.io .

We present a method enabling the scaling of NeRFs to learn a large number of semantically-similar scenes. We combine two techniques to improve the required training time and memory cost per scene. First, we learn a 3D-aware latent space in which we train Tri-Plane scene representations, hence reducing the resolution at which scenes are learned. Moreover, we present a way to share common information across scenes, hence allowing for a reduction of model complexity to learn a particular scene. Our method reduces effective per-scene memory costs by 44% and per-scene time costs by 86% when training 1000 scenes. Our project page can be found at https://3da-ae.github.io ., Comment: Camera-ready version accepted at 3DMV-CVPR 2024

Modeling large scenes from unconstrained images has proven to be a major challenge in computer vision. Existing methods tackling in-the-wild scene modeling operate in closed-world settings, where no conditioning on priors acquired from real-world images is present. We propose RefinedFields, which is, to the best of our knowledge, the first method leveraging pre-trained models to improve in-the-wild scene modeling. We employ pre-trained networks to refine K-Planes representations via optimization guidance using an alternating training procedure. We carry out extensive experiments and verify the merit of our method on synthetic data and real tourism photo collections. RefinedFields enhances rendered scenes with richer details and improves upon its base representation on the task of novel view synthesis in the wild. Our project page can be found at https://refinedfields.github.io., Comment: Corrected Table 2, where some comparisons were done among models trained at different resolutions

We demonstrate and analyse a novel approach to enhance the threshold power of stimulated Brillouin scattering (SBS) in optical fibers, using a longitudinal compressive strain gradient. We derive analytical expressions for the power spectral density of the backscattered Stokes wave in the general case of passive and amplifying optical fibers, by considering the strain and optical power distributions. Our method provides an accurate prediction of the SBS gain spectrum, that we illustrate with a quantitative comparison between measurements and calculations of the SBS Stokes spectra, before and after applying the compression gradient. Our experimental results demonstrate the successful enhancement of the SBS threshold power by a factor of about 3 for the passive fiber and 2 for the amplifying fiber. The enhancement that we manage to calculate in the case of the passive fiber is in perfect agreement with the experimental result.

We developed a new holmium-doped triple-clad fiber (Ho-3CF), reducing the clad diameter to improve the overlap between pump and doped-core, and the holmium concentration to reduce the influence of ion clustering on the laser efficiency. We illustrate here the performance of this fiber in a laser oscillator configuration. The laser emission is centered at 2.12 $\mu$m by a fiber Bragg grating, and the active fiber is pumped in the clad by a 1.94 $\mu$m thulium-doped fiber laser. The slope efficiency reaches 73% with respect to absorbed power (60% considering coupled power), for a maximum signal power of 62 W delivered on a quasi single spatial mode (M$^2$=1.2). To the best of our knowledge, this slope efficiency is the highest reported for a high power clad-pumped fiber laser emitting at a wavelength higher than 2.1 $\mu$m. We also analyze experimentally the impact of the pump Numerical Aperture (NA), at the input of the Ho-3CF, on the laser efficiency. Finally, we use our numerical simulation to comment on the best choice of pump wavelength.

Location retrieval based on visual information is to retrieve the location of an agent (e.g. human, robot) or the area they see by comparing the observations with a certain form of representation of the environment. Existing methods generally require precise measurement and storage of the observed environment features, which may not always be robust due to the change of season, viewpoint, occlusion, etc. They are also challenging to scale up and may not be applicable for humans due to the lack of measuring/imaging devices. Considering that humans often use less precise but easily produced qualitative spatial language and high-level semantic landmarks when describing an environment, a qualitative location retrieval method is proposed in this work by describing locations/places using qualitative place signatures (QPS), defined as the perceived spatial relations between ordered pairs of co-visible landmarks from viewers' perspective. After dividing the space into place cells each with individual signatures attached, a coarse-to-fine location retrieval method is proposed to efficiently identify the possible location(s) of viewers based on their qualitative observations. The usability and effectiveness of the proposed method were evaluated using openly available landmark datasets, together with simulated observations by considering the possible perception error.

With impressive results in applications relying on feature learning, deep learning has also blurred the line between algorithm and data. Pick a training dataset, pick a backbone network for feature extraction, and voil\`a ; this usually works for a variety of use cases. But the underlying hypothesis that there exists a training dataset matching the use case is not always met. Moreover, the demand for interconnections regardless of the variations of the content calls for increasing generalization and robustness in features. An interesting application characterized by these problematics is the connection of historical and cultural databases of images. Through the seemingly simple task of instance retrieval, we propose to show that it is not trivial to pick features responding well to a panel of variations and semantic content. Introducing a new enhanced version of the Alegoria benchmark, we compare descriptors using the detailed annotations. We further give insights about the core problems in instance retrieval, testing four state-of-the-art additional techniques to increase performance.

Advances in high resolution remote sensing image analysis are currently hampered by the difficulty of gathering enough annotated data for training deep learning methods, giving rise to a variety of small datasets and associated dataset-specific methods. Moreover, typical tasks such as classification and retrieval lack a systematic evaluation on standard benchmarks and training datasets, which make it hard to identify durable and generalizable scientific contributions. We aim at unifying remote sensing image retrieval and classification with a new large-scale training and testing dataset, SF300, including both vertical and oblique aerial images and made available to the research community, and an associated fine-tuning method. We additionally propose a new adversarial fine-tuning method for global descriptors. We show that our framework systematically achieves a boost of retrieval and classification performance on nine different datasets compared to an ImageNet pretrained baseline, with currently no other method to compare to., Comment: Performance margin with the proposed method is not statistically significant. Please refer to http://alegoria.ign.fr/en/SF300_dataset if you are interested in the dataset

We report on the realization of an all-fiber laser source that delivers single-frequency pulses at 1645 nm, on a linearly polarized single-mode beam, based on stimulated Raman scattering in passive fibers. The pulse energy reaches 14 $\mu$J, for a repetition rate of 20 kHz, and the spectral linewidth is 9.5 MHz, almost Fourier-transform limited for the 100 ns square pulses. To the best of our knowledge, this energy is the highest reported at 1645 nm in an all-fiber laser source. Our method consists in reducing the stimulated Brillouin scattering (SBS) gain for the pump and signal pulses, respectively by sweeping the optical frequency of the pump beam, and by applying a strain gradient on the amplification fiber. This compact laser source is now used in a transportable lidar system to measure simultaneously wind velocity and methane (CH4) concentration.

The study of records in the Linear Drift Model (LDM) has attracted much attention recently due to applications in several fields. In the present paper we study $\delta$-records in the LDM, defined as observations which are greater than all previous observations, plus a fixed real quantity $\delta$. We give analytical properties of the probability of $\delta$-records and study the correlation between $\delta$-record events. We also analyse the asymptotic behaviour of the number of $\delta$-records among the first $n$ observations and give conditions for convergence to the Gaussian distribution. As a consequence of our results, we solve a conjecture posed in J. Stat. Mech. 2010, P10013, regarding the total number of records in a LDM with negative drift. Examples of application to particular distributions, such as Gumbel or Pareto are also provided. We illustrate our results with a real data set of summer temperatures in Spain, where the LDM is consistent with the global-warming phenomenon., Comment: 30 pages, 12 figures

We investigate the decoherence of Er$^{3+}$ in Y$_2$SiO$_5$ at low magnetic fields using the photon-echo technique. We reproduce accurately a variety of the decay curves with a unique coherence time by considering the so-called superhyperfine modulation induced by a large number of neighbouring spins. There is no need to invoke any characteristic time of the spin fluctuations to reproduce very different decay curves. The number of involved nuclei increases when the magnetic is lowered. The experiment is compared with a model associating 100 surrounding ions with their exact positions in the crystal frame. We also derive an approximate spherical model (angular averaging) to interpret the main feature the observed decay curves close to zero-field.

Visual localization is a useful alternative to standard localization techniques. It works by utilizing cameras. In a typical scenario, features are extracted from captured images and compared with geo-referenced databases. Location information is then inferred from the matching results. Conventional schemes mainly use low-level visual features. These approaches offer good accuracy but suffer from scalability issues. In order to assist localization in large urban areas, this work explores a different path by utilizing high-level semantic information. It is found that object information in a street view can facilitate localization. A novel descriptor scheme called "semantic signature" is proposed to summarize this information. A semantic signature consists of type and angle information of visible objects at a spatial location. Several metrics and protocols are proposed for signature comparison and retrieval. They illustrate different trade-offs between accuracy and complexity. Extensive simulation results confirm the potential of the proposed scheme in large-scale applications. This paper is an extended version of a conference paper in CBMI'18. A more efficient retrieval protocol is presented with additional experiment results., Comment: 12 pages, 22 figures, Multimedia Tools and Applications (2020)

We report on our realization of a high power holmium doped fiber laser, together with the validation of our numerical simulation of the laser. We first present the rare absolute measurements of the physical parameters that are mandatory to model accurately the laser-holmium interactions in our silica fiber. We then describe the realization of the clad-pumped laser, based on a triple-clad large mode area holmium (Ho) doped silica fiber. The output signal power is 90 W at 2120 nm, with an efficiency of about 50% with respect to the coupled pump power. This efficiency corresponds to the state of the art for clad-pumped Ho-doped fiber lasers in the 100 W power class. By comparing the experimental results to our simulation, we demonstrate its validity, and use it to show that the efficiency is limited, for our fiber, by the non-saturable absorption caused by pair induced quenching between adjacent holmium ions., Comment: 14 pages, 7 figures

This article proposes to study the behavior of recent and efficient state-of-the-art deep-learning based image descriptors for content-based image retrieval, facing a panel of complex variations appearing in heterogeneous image datasets, in particular in cultural collections that may involve multi-source, multi-date and multi-view Permission to make digital, Comment: SUMAC '19, 2019

Consider a number, finite or not, of urns each with fixed capacity $r$ and balls randomly distributed among them. An overflow is the number of balls that are assigned to urns that already contain $r$ balls. When $r=1$, using analytic methods, Hwang and Janson gave conditions under which the overflow (which in this case is just the number of balls landing in non--empty urns) has an asymptotically Poisson distribution as the number of balls grows to infinity. Our aim here is to systematically study the asymptotics of the overflow in general situation, i.~e. for arbitrary $r$. In particular, we provide sufficient conditions for both Poissonian and normal asymptotics for general $r$, thus extending Hwang--Janson's work. Our approach relies on purely probabilistic methods., Comment: 4figures

We investigate the erbium flip-flop dynamics as a limiting factor of the electron spin lifetime and more generally as an indirect source of decoherence in rare-earth doped insulators. Despite the random isotropic arrangement of dopants in the host crystal, the dipolar interaction strongly depends on the magnetic field orientation following the strong anisotropy of the $g$-factor. In Er$^{3+}$:Y$_2$SiO$_5$, we observe by transient optical spectroscopy a three orders of magnitude variation of the erbium flip-flop rate (10ppm dopant concentration). The measurements in two different samples, with 10ppm and 50ppm concentrations, are well-supported by our analytic modeling of the dipolar coupling between identical spins with an anisotropic $g$-tensor. The model can be applied to other rare-earth doped materials. We extrapolate the calculation to Er$^{3+}$:CaWO$_4$, Er$^{3+}$:LiNbO$_3$ and Nd$^{3+}$:Y$_2$SiO$_5$ at different concentrations.

The influence of low temperatures on the performance of a high-power single-frequency fiber laser amplifier is evaluated with a numerical simulation. Cooling the fiber can allow to take advantage of both higher damping of the acoustic waves in the silica glass, and higher laser efficiency. We first report on the measurement of the stimulated Brillouin scattering (SBS) threshold in a silica fiber as a function of the temperature from 300~K down to 77~K. We then present the measurements of small-signal absorption and gain spectra of an erbium doped alumino-phosphosilicate fiber, at room temperature and liquid nitrogen temperature. Based on these data, we derive a numerical study of the combined effects of cooling on the SBS threshold and the amplifier efficiency, and conclude on the interest of this technique for SBS limited high power Er doped fiber amplifiers (EDFA). The temperature increase caused by the pump laser in the fiber core is also addressed.

We study photon echo generation in disordered media with the help of multiple scattering theory based on diagrammatic approach and numerical simulations. We show that a strong correlation exists between the driving fields at the origin of the echo and the echo beam. Opening the way to a better understanding of non-linear wave propagation in complex materials, this work supports recent experimental results with applications to the measurement of the optical dipole lifetime $T_2$ in powders.

In Er$^{3+}$:Y$_2$SiO$_5$, we demonstrate the selective optical addressing of the $^{89}$Y$^{3+}$ nuclear spins through their superhyperfine coupling with the Er$^{3+}$ electronic spins possessing large Land\'e $g$-factors. We experimentally probe the electron-nuclear spin mixing with photon echo techniques and validate our model. The site-selective optical addressing of the Y$^{3+}$ nuclear spins is designed by adjusting the magnetic field strength and orientation. This constitutes an important step towards the realization of long-lived solid-state qubits optically addressed by telecom photons., Comment: 5 pages, 4 figures, supplementary material (3 pages)

This paper is focused on the statistical analysis of probability measures $\nu_{1},\ldots,\nu_{n}$ on $\mathbb{R}$ that can be viewed as independent realizations of an underlying stochastic process. We consider the situation of practical importance where the random measures $\nu_{i}$ are absolutely continuous with densities $f_{i}$ that are not directly observable. In this case, instead of the densities, we have access to datasets of real random variables $(X_{i,j})_{1 \leq i \leq n; \; 1 \leq j \leq p_{i} }$ organized in the form of $n$ experimental units, such that $X_{i,1},\ldots,X_{i,p_{i}}$ are iid observations sampled from a random measure $\nu_{i}$ for each $1 \leq i \leq n$. In this setting, we focus on first-order statistics methods for estimating, from such data, a meaningful structural mean measure. For the purpose of taking into account phase and amplitude variations in the observations, we argue that the notion of Wasserstein barycenter is a relevant tool. The main contribution of this paper is to characterize the rate of convergence of a (possibly smoothed) empirical Wasserstein barycenter towards its population counterpart in the asymptotic setting where both $n$ and $\min_{1 \leq i \leq n} p_{i}$ may go to infinity. The optimality of this procedure is discussed from the minimax point of view with respect to the Wasserstein metric. We also highlight the connection between our approach and the curve registration problem in statistics. Some numerical experiments are used to illustrate the results of the paper on the convergence rate of empirical Wasserstein barycenters.

We investigate cross-relaxation interactions between Tm and Al in Tm:YAG using two optical methods: spectral holeburning and stimulated echoes. These interactions lead to a reduction in the hyperfine lifetime at magnetic fields that bring the Tm hyperfine transition into resonance with an Al transition. We develop models for measured echo decay curves and holeburning spectra near a resonance, which are used to show that the Tm-Al interaction has a resonance width of 10~kHz and reduces the hyperfine lifetime to 0.5 ms. The antihole structure is consistent with an interaction dominated by the Al nearest neighbors at 3.0 Angstroms, with some contribution from the next nearest neighbors at 3.6 Angstroms., Comment: 13 pages, 9 figures

We study the bandwidth and multiplexing capacity of an erbium-doped optical memory for quantum storage purposes. We concentrate on the protocol ROSE (Revival of a Silenced Echo) because it has the largest potential multiplexing capacity. Our analysis is applicable to other protocols that involve strong optical excitation. We show that the memory performance is limited by instantaneous spectral diffusion and we describe how this effect can be minimised to achieve optimal performance.

Interlaced Spin Grating is a scheme for the preparation of spectro-spatial periodic absorption gratings in a inhomogeneously broadened absorption profile. It relies on the optical pumping of atoms in a nearby long-lived ground state sublevel. The scheme takes advantage of the sublevel proximity to build large contrast gratings with unlimited bandwidth and preserved average optical depth. It is particularly suited to Tm-doped crystals in the context of classical and quantum signal processing. In this paper, we study the optical pumping dynamics at play in an Interlaced Spin Grating and describe the corresponding absorption profile shape in an optically thick atomic ensemble. We show that, in Tm:YAG, the diffraction efficiency of such a grating can reach 18.3% in the small angle, and 11.6% in the large angle configuration when the excitation is made of simple pulse pairs, considerably outperforming conventional gratings., Comment: 11 pages, 13 figures in Physical Review A, 2015

We implement the ROSE protocol in an erbium doped solid, compatible with the telecom range. The ROSE scheme is an adaptation of the standard 2-pulse photon echo to make it suitable for a quantum memory. We observe an efficiency of 40% in a forward direction by using specific orientations of the light polarizations, magnetic field and crystal axes.

To store and retrieve signals at the single photon level, various photon echo schemes have resorted to complex preparation steps involving ancillary shelving states in multi-level atoms. For the first time, we experimentally demonstrate photon echo operation at such a low signal intensity without any preparation step, which allows us to work with mere two-level atoms. This simplified approach relies on the so-coined "Revival Of Silenced Echo" (ROSE) scheme. Low noise conditions are obtained by returning the atoms to the ground state before the echo emission. In the present paper we manage ROSE in photon counting conditions, showing that very strong control fields can be compatible with extremely weak signals, making ROSE consistent with quantum memory requirements., Comment: 10 pages, 5 figures

We introduce the method of Geodesic Principal Component Analysis (GPCA) on the space of probability measures on the line, with finite second moment, endowed with the Wasserstein metric. We discuss the advantages of this approach, over a standard functional PCA of probability densities in the Hilbert space of square-integrable functions. We establish the consistency of the method by showing that the empirical GPCA converges to its population counterpart, as the sample size tends to infinity. A key property in the study of GPCA is the isometry between the Wasserstein space and a closed convex subset of the space of square-integrable functions, with respect to an appropriate measure. Therefore, we consider the general problem of PCA in a closed convex subset of a separable Hilbert space, which serves as basis for the analysis of GPCA and also has interest in its own right. We provide illustrative examples on simple statistical models, to show the benefits of this approach for data analysis. The method is also applied to a real dataset of population pyramids.

Based on the similarity of paraxial diffraction and dispersion mathematical descriptions, the temporal imaging of optical pulses combines linear dispersive filters and quadratic phase modulations operating as time lenses. We consider programming a dispersive filter near atomic resonance in rare earth ion doped crystals, which leads to unprecedented high values of dispersive power. This filter is used in an approximate imaging scheme, combining a single time lens and a single dispersive section and operating as a time reversing device, with potential applications in radio-frequency signal processing. This scheme is closely related to three-pulse photon echo with chirped pulses but the connection with temporal imaging and dispersive filtering emphasizes new features., Comment: 21 pages, 11 figures