Christophe Ducottet's research while affiliated with French National Centre for Scientific Research and other places

... For instance, AI-enabled nanoparticle synthesis platforms and nanoparticle delivery systems (AI-assisted algorithms) have been adopted to optimize the synthesis and delivery of magnetic nanoparticles for anticancer drug delivery, respectively. It uses machine learning, deep learning, and computer vision techniques to predict the optimal parameters for nanoparticle synthesis that accurately target the nanoparticles to the desired cells [258,259]. Furthermore, AI-assisted algorithms have also been employed to track the magnetic nanoparticles and monitor the toxicity of the drug during delivery to provide insights into cancer progression and identify potential therapeutic strategies [260]. It uses natural language processing, deep learning, and computer vision techniques to accurately detect the nanoparticles at the specified locations (tumor sites) and assess the levels of toxicity of the drug, respectively. ...

... In fact, meaningful results were obtained just with U-Net both in liquid and gas phases, and under temperature changes, surpassing the former traditional and unsupervised approaches. [174][175][176] In any case, it is clear that ML has much more to say in in situ EM, and it will surely explode as soon as the in situ machinery becomes a much wider standard within the community. For instance, we envision the use of CS to reduce the required frame rate, similarly to CS-tomography, further allowing the chemical and physical tracking of beam-sensitive materials. ...

... To address the issue of missing inter-slice information and bridge the gap further between 2D and 3D strategies, a distance transform was applied to the initial volume before slicing. This approach, introduced in [15], allows to enhance 2D images by adding geodesic information in non-textured areas. The ADS-net, trained on augmented 2D images by 3D geodesic information, can be considered somehow as a 2.5D network. ...

... Then we investigated the problem from a domain adaptive perspective to assess the ability of object detector trained on a chairlift to CHAPTER 2. STATE OF THE ART perform well on new unseen chairlift [37]. While in [38] we opted to exploit the geometrical constraint by guiding shallow and deep classifiers using binary masks representing the safety bar, then we proposed a solution to boost the generalizability of the deeper classifier. In the later sections of this chapter, we present an overview of the available literature on each of these aspects. ...

... Furthermore, SCL [30] learns more discriminative representations by interacting different losses and training strategy. Recently, Alqasir et al. indicated that adapting the region proposal sub-network in Faster R-CNN is crucial [31]. Existing works strive into diminishing the domain gap in both global feature (or image) level and local feature (or instance) level. ...

... Although dilation/contraction processes were not studied in our experiments, the decrease in porosity may suggest that contraction of the bed is likely the only process that occurred. As explained by Frey et al. (2020), a moving bedload layer creates space for particles to infiltrate and, therefore, decreasing the porosity. However, we believe that the rising limb of the hydrograph of short time-to-peak floods might cause dilation during under-threshold flows. ...

... The corresponding roughness parameters are = 1.180 mm, = 2.25 mm −2 and = 0.0419 mm. Let us now discuss the physical relevance of such a coarse optimal resolution, based on the Power Spectrum Density (PSD) of the same rough topography given in Fig. S1 of Sahli et al. (2019). The PSD has a plateau at large scales followed by a power law at smaller scales. ...

... The opacity of the intruder among transparent beads combined with the high frequency recording allowed us to follow the position of the large particle frame by frame. A Particle Tracking Velocimetry (PTV) algorithm developped by Lafaye de Micheaux et al. (2018) was used. This algorithm relies on two steps. ...

... Thus, its detection is challenging. Since the proposed algorithm detects symmetry axes as primary lines, we compare the proposed algorithm with the existing methods [8,9,14,33] Fig. 10: Detection results of symmetry axes. The ground-truth axes are in red, the detection results of Loy and Eklundh [33] are in green, and those of the proposed algorithm are in yellow. ...

... Marcelo et al. [29] outperformed other competitors in the 2017 competition with a registration technique that registered the collection of original points to their mirror-reflected counterparts. Methods based on image gradient [30] or important edges also achieved robust symmetry axes extraction [31][32][33]. ...