arXiv subjects

Pierre Lemaire

Publications and source records attributed to Pierre Lemaire.

7 recordsLinked to original sources

arXiv2026

Curriculum-guided Change Detection Training: Toward Accurate Serac Fall Monitoring

Change Detection (CD) aims to identify semantic or structural changes from nearly registered multi-temporal images. While recent advances in training methodologies have largely focused on semi-supervised learning and consistency regularization, alternative training paradigms remain underexplored. In particular, most deep CD methods rely on uniform sampling during training, implicitly assuming that all training samples contribute equally to the optimization process. However, such naive sampling can introduce noisy gradients and hinder robust representation learning. To address this limitation, we propose a curriculum learning framework tailored for change detection. Our approach investigates two complementary difficulty measures: the Solar Angular Gap (SAG), a physically grounded proxy for acquisition-condition variability, and the Structural Similarity Index Measure (SSIM), which evaluates appearance similarity between image pairs. Based on these criteria, the framework progressively introduces challenging samples during training, enabling models to learn robust representations in a coarse-to-fine manner. We evaluate our method on the challenging SeracFallDet benchmark, where results demonstrate consistent improvements of the proposed approach over standard uniform-sampling strategies for both pixel-based and object-based approaches. These results highlight the potential of curriculum learning to improve robustness in deep change detection. Importantly, our training framework is orthogonal to existing CD architectures, making it readily applicable to a broad range of methods.

cs.CV↗

arXiv2026

Scheduling of Star Observations under Uncertain Conditions: A Comparison of Models and Solvers

We consider a scheduling problem for star observations using a telescope. Identical nights are available to observe stars that have been identified as of interest by astronomers. Each star yields a profit if it is observed; in that case, the observation must respect a visibility window and a minimum observation duration. The objective is to maximize the total profit of the observations that are actually performed. By interpreting nights as machines and stars as jobs to be processed, this problem is a classic scheduling setting of parallel machines. In the practical star-observation problem, the ability to observe depends on meteorological and atmospheric conditions, which are not known when the schedule is computed. A possible modeling is to consider that each night can be either perfect and everything is observable, either terrible and nothing is observable. The challenge is therefore to propose a ''high-quality'' schedule without knowing the actual number of nights (machines) that will ultimately be available. We focus here on evaluating the worst case within the robust optimization paradigm.

math.OC↗

arXiv2026

Revisiting Change Detection Methods for their Application to Serac Fall Time-Lapse Monitoring

In an era where climate change aggravates environmental uncertainties, the identification and detection of event precursors are becoming crucial to mitigate the impacts of disastrous natural hazards. While classical sensors such as interferometric lasers or seismometers are reliable, their widespread deployment is often hindered by logistical and economic barriers, leaving numerous blind spots. Time-lapse cameras, which already provide cost-effective, high-resolution visual context to such sensors, present a promising alternative. However, processing their output automatically faces significant challenges, notably linked to extreme shape and lighting variations. Overcoming those issues is essential to deploy them at large-scale as a monitoring tool. This paper introduces a novel sub-task of change detection, namely volumetric change detection, applied to time-lapse cameras and slope instabilities. We conduct a comprehensive review of state-of-the-art change detection methods and related tasks, analyze their core components and assess their applicability to this context. To that end, we introduce the new dataset SeracFallDet, which contains serac fall annotations and has been thoroughly annotated to meet the latter demand. Through generalization experiments, we demonstrate that dense and semi-dense feature matching, although not trained specifically for this task, exhibit robust performance. Alternatively, supervised approaches struggle with data scarcity and annotation imbalance. This suggests that hybrid methods may offer a path forward by leveraging the strengths of both tasks. These findings highlight the potential of feature matching techniques and the need for further innovation to overcome the challenges of real-world deployment in environmental monitoring.

cs.CV↗

arXiv2026

Scheduling Ground-Based Telescope Observations with Uncertain Nights

The observation of celestial objects is a fundamental activity in astronomy. Ground-based and space telescopes are used to gather electromagnetic radiation from space, allowing astronomers to study a wide range of celestial objects and phenomena, such as stars, planets, galaxies, and black holes. The European Southern Observatory (ESO) charges each night 83 kEUR (Milli et al. 2019), so the schedules of the telescopes are really important in order to optimize every second. Ground-based telescopes are affected by meteorological conditions, such as clouds, wind, and atmospheric turbulence. Accurate scheduling of observations in the presence of such uncertainties can significantly improve the efficiency of telescopes use and support from automated tools is highly desirable. In this paper, we study a mathematical approach for scheduling ground-based telescope observations under an uncertain number of clear nights due to uncertain weather and atmospheric conditions. The model considers multiple targets, uncertain number of nights, and various observing constraints. We demonstrate the viability and effectiveness of an approach based on stochastic optimization and reactive strategy comparing it against other methods.

math.OC↗

arXiv2025

Planification d'observations sur t{é}lescope avec nuits incertaines

astro-ph.IM↗

arXiv2024

Early Detection of Critical Urban Events using Mobile Phone Network Data

Network Signalling Data (NSD) have the potential to provide continuous spatio-temporal information about the presence, mobility, and usage patterns of cell phone services by individuals. Such information is invaluable for monitoring large urban areas and supporting the implementation of decision-making services. When analyzed in real time, NSD can enable the early detection of critical urban events, including fires, large accidents, stampedes, terrorist attacks, and sports and leisure gatherings, especially if these events significantly impact mobile phone network activity in the affected areas. This paper presents empirical evidence that advanced NSD can detect anomalies in mobile traffic service consumption, attributable to critical urban events, with fine spatial and temporal resolutions. We introduce two methodologies for real-time anomaly detection from multivariate time series extracted from large-scale NSD, utilizing a range of algorithms adapted from the state-of-the-art in unsupervised machine learning techniques for anomaly detection. Our research includes a comprehensive quantitative evaluation of these algorithms on a large-scale dataset of NSD service consumption for the Paris region. The evaluation uses an original dataset of documented critical or unusual urban events. This dataset has been built as a ground truth basis for assessing the algorithms performance. The obtained results demonstrate that our framework can detect unusual events almost instantaneously and locate the affected areas with high precision, largely outperforming random classifiers. This efficiency and effectiveness underline the potential of NSD-based anomaly detection in significantly enhancing emergency response strategies and urban planning.

cs.CY↗

arXiv2017

A study of variability induced by events dependency in microelectronic production

-Complex manufacturing systems are subject to high levels of variability that decrease productivity, increase cycle times and severely impact the systems tractability. As accurate modelling of the sources of variability is a cornerstone to intelligent decision making, we investigate the consequences of the assumption of independent and identically distributed variables that is often made when modelling sources of variability such as down-times, arrivals, or process-times. We first explain the experiment setting that allows, through simulations and statistical tests, to measure the variability potential stored in a specific sequence of data. We show from industrial data that dependent behaviors might actually be the rule with potentially considerable consequences in terms of cycle time. As complex industries require strong levers to allow their tractability, this work underlines the need for a richer and more accurate modelling of real systems. Keywords-variability; cycle time; dependent events; simulation; complex manufacturing; industry 4.0 I. Accurate modelling of variability and the independence assumption Industry 4.0 is said to be the next industrial revolution. The proper use of real-time information in complex manufacturing systems is expected to allow more customization of products in highly flexible production factories. Semiconductor High Mix Low Volume (HMLV) manufacturing facilities (called fabs) are one example of candidates for this transition towards "smart industries". However, because of the high levels of variability, the environment of a HMLV fab is highly stochastic and difficult to manage. The uncontrolled variability limits the predictability of the system and thus the ability to meet delivery requirements in terms of volumes, cycle times and due dates. Typically, the HMLV STMicroelectronics Crolles 300 fab regularly experiences significant mix changes that result in unanticipated bottlenecks, leading to firefighting to meet commitment to customers. The overarching goal of our strategy is to improve the forecasting of future occurrences of bottlenecks and cycle time issues in order to anticipate them through allocation of the correct attention and resources. Our current finite capacity projection engine can effectively forecast bottlenecks, but it does not include reliable cycle time estimates. In order to enhance our projections, better forecast cycle time losses (queuing times), improve the tractability of our system and reduce our cycle times, we now need accurate dynamic cycle time predictions. As increased cycle-time is the main reason workflow variability is studied (both by the scientific community and practitioners, see e.g. [1] and [2]), what follows concentrates on cycle times. Moreover, the "variability" we account for should be understood as the potential to create higher cycle times, even though "variability" may be understood in a broader meaning. This choice is made for the sake of clarity, but the methodology we propose and the discussion we lead can be applied to any other measurable indicator. Sources of variability have been intensely investigated in both the literature and the industry, and tool down-times, arrivals variability as well as process-time variability are recognized as the major sources of variability in that sense that they create higher cycle times (see [3] for a review and discussion). As a consequence, these factors are widely integrated into queuing formulas and simulation models with the objective to better model the complex reality of manufacturing facilities. One commonly accepted assumption in the development of these models is that the variables (MTBF, MTTR, processing times, time between arrivals, etc.) are independent and identically distributed (i.i.d.) random variables. However, these assumptions might be the reason for models inaccuracies as [4] points out in a literature review on queuing theory. Several authors have studied the potential effects of dependencies, such as [5] who studied the potential effects of dependencies between arrivals and process-times or [6] who investigated dependent process times, [4] also gives further references for studies on dependencies effects. In a previous work [3], we pinpointed a few elements from industrial data that questioned the viability of this assumption in complex manufacturing systems. Figure 1: Number of arrivals per week from real data (A) and generated by removing dependencies (B)

stat.AP↗