Superpoint-based pipelines provide an efficient alternative to point- or voxel-based 3D semantic segmentation, but are often bottlenecked by their CPU-bound partition step. We propose a learnable, fully GPU partitioning algorithm that generates geometrically and semantically coherent superpoints 13× faster than prior methods. Our module is compact (under 60k parameters), trains in under 20 minutes with a differentiable surrogate loss, and requires no handcrafted features. Combined with a lightweight superpoint classifier, the full pipeline fits in < 2 MB of VRAM, scales to multi-million-point scenes, and supports real-time inference. With 72× faster inference and 120× fewer parameters, EZ-SP matches the accuracy of point-based SOTA models across three domains: indoor scans (S3DIS), autonomous driving (KITTI-360), and aerial LiDAR (DALES). Code and pretrained models are accessible at github.com/drprojects/superpoint_transformer.

We introduce a highly efficient method for panoptic segmentation of large 3D point clouds by redefining this task as a scalable graph clustering problem. This approach can be trained using only local auxiliary tasks, thereby eliminating the resource-intensive instance-matching step during training. Moreover, our formulation can easily be adapted to the superpoint paradigm, further increasing its efficiency. This allows our model to process scenes with millions of points and thousands of objects in a single inference. Our method, called SuperCluster, achieves a new state-of-the-art panoptic segmentation performance for two indoor scanning datasets: $50.1$ PQ ($+7.8$) for S3DIS Area~5, and $58.7$ PQ ($+25.2$) for ScanNetV2. We also set the first state-of-the-art for two large-scale mobile mapping benchmarks: KITTI-360 and DALES. With only $209$k parameters, our model is over $30$ times smaller than the best-competing method and trains up to $15$ times faster. Our code and pretrained models are available at https://github.com/drprojects/superpoint_transformer.

We introduce a novel superpoint-based transformer architecture for efficient semantic segmentation of large-scale 3D scenes. Our method incorporates a fast algorithm to partition point clouds into a hierarchical superpoint structure, which makes our preprocessing 7 times faster than existing superpoint-based approaches. Additionally, we leverage a self-attention mechanism to capture the relationships between superpoints at multiple scales, leading to state-of-the-art performance on three challenging benchmark datasets: S3DIS (76.0% mIoU 6-fold validation), KITTI-360 (63.5% on Val), and DALES (79.6%). With only 212k parameters, our approach is up to 200 times more compact than other state-of-the-art models while maintaining similar performance. Furthermore, our model can be trained on a single GPU in 3 hours for a fold of the S3DIS dataset, which is 7x to 70x fewer GPU-hours than the best-performing methods. Our code and models are accessible at github.com/drprojects/superpoint_transformer.

Recent works on 3D semantic segmentation propose to exploit the synergy between images and point clouds by processing each modality with a dedicated network and projecting learned 2D features onto 3D points. Merging large-scale point clouds and images raises several challenges, such as constructing a mapping between points and pixels, and aggregating features between multiple views. Current methods require mesh reconstruction or specialized sensors to recover occlusions, and use heuristics to select and aggregate available images. In contrast, we propose an end-to-end trainable multi-view aggregation model leveraging the viewing conditions of 3D points to merge features from images taken at arbitrary positions. Our method can combine standard 2D and 3D networks and outperforms both 3D models operating on colorized point clouds and hybrid 2D/3D networks without requiring colorization, meshing, or true depth maps. We set a new state-of-the-art for large-scale indoor/outdoor semantic segmentation on S3DIS (74.7 mIoU 6-Fold) and on KITTI-360 (58.3 mIoU). Our full pipeline is accessible at https://github.com/drprojects/DeepViewAgg, and only requires raw 3D scans and a set of images and poses.

During the last decade, the development of nanomaterials to penetrate inside living cells has been the focus of a large number of studies, with applications for the biomedical field. However, the further dynamics of these nanomaterials inside the cells is dictated by the intracellular environment and in particular its mechanical properties. The mechanical characteristics of the cell interior can be probed with either active or passive microrheological approaches. However, active intracellular microrheology is still in its infancy, owing to the difficulty of inserting probes that can be manipulated by external forces. Here we review recent active microrheology studies using magnetic nanoprobes inserted into endosomes or phagosomes as useful approaches for measuring frequency-dependent viscoelasticity, for mapping the viscoelastic landscape, as well as for identifying the contribution of individual cytoskeleton components and the influence of cell motility. The results of such direct measurements challenge the validity of more typical passive approaches in which the spontaneous displacement of embedded nanoprobes is measured. Here we discuss that one must distinguish probes suitable for use in conditions of thermal equilibrium, whose movements reflect the mechanical environment from probes that interact actively with the cytoplasm and cytoskeleton, in a state of nonequilibrium for which fluctuation-dissipation theorem no longer holds. However, when data on these probes' viscoelastic microenvironment is available, such passive probe movements can yield useful information on the forces responsible for intracellular activity.

Magnetically labelled cells are finding a wealth of applications for in vitro analysis as well as in vivo treatments. Sorting of cells into subpopulations based on their magnetite loading is an important step in such procedures. Here, we study the sorting of monocytes and macrophages which internalise nanoparticles to different extents based on their endocytotic capacity. Macrophages featured a high endocytotic activity and were found to internalise between 4 and 60 pg of iron per cell. They were successfully sorted into five subpopulations of narrow iron loading distributions via on-chip free-flow magnetophoresis, thus demonstrating the potential of sorting of relatively similarly loaded cells. Monocytes featured a low endocytotic capacity and took on 1 to 4 pg of iron per cell. Mixtures of monocytes and macrophages were successfully sorted within the free-flow magnetophoresis chip and good purity (> 88%), efficacy (> 60%) and throughput (from 10 to 100 cells s(-1)) could be achieved. The introduced method constitutes a viable tool for studies of endocytotic capacity and sorting/selection of cells based on this functionality.

Background: Regulation of intracellular trafficking is a central issue in cell biology. The forces acting on intracellular vesicles (endosomes) can be assessed in living cells by using a combination of active and passive microrheology.Methodology/Principal Findings: This dual approach is based on endosome labeling with magnetic nanoparticles. The resulting magnetic endosomes act both as probes that can be manipulated with external magnetic fields to infer the viscoelastic modulus of their surrounding microenvironment, and as biological vehicles that are trafficked along the microtubule network by means of forces generated by molecular motors. The intracellular viscoelastic modulus exhibits power law dependence with frequency, which is microtubule and actin-dependent. The mean square displacements of endosomes do not follow the predictions of the fluctuation-dissipation theorem, which offers evidence for active force generation. Microtubule disruption brings the intracellular medium closer to thermal equilibrium: active forces acting on the endosomes depend on microtubule-associated motors. The power spectra of these active forces, deduced through the use of a generalized Langevin equation, show a power law decrease with frequency and reveal an actin-dependent persistence of the force with time. Experimental spectra have been reproduced by a simple model consisting in a series of force steps power-law distributed in time. This model enlightens the role of the cytoskeleton dependent force exerted on endosomes to perform intracellular trafficking.Conclusions/Significance: In this work, the influence of cytoskeleton components and molecular motors on intracellular viscoelasticity and transport is addressed. The use of an original probe, the magnetic endosome, allows retrieving the power spectrum of active forces on these organelles thanks to interrelated active and passive measures. Finally a computational model gives estimates of the force itself and hence of the number of the motors pulling on endosomes.

We present temperature control as a simple yet highly efficient means of improving particle separations in on-chip free-flow magnetophoresis. Increased temperatures resulted in a decrease in solution viscosity, which in turn yielded greater particle deflection distances and increased the separation resolution of two particle populations. This inexpensive and effective technique would also be of great benefit to other continuous flow processes involving the deflection of particles.

The in vitro generation of engineered tissue constructs involves the seeding of cells into porous scaffolds. Ongoing challenges are to design scaffolds to meet biochemical and mechanical requirements and to optimize cell seeding in the constructs. In this context, we have developed a simple method based on a magnetic tweezer set-up to manipulate, probe, and position magnetic objects inside a porous scaffold. The magnetic force acting on magnetic objects of various sizes serves as a control parameter to retrieve the local viscosity of the scaffolds internal channels as well as the stiffness of the scaffolds pores. Labeling of human stem cells with iron oxide magnetic nanoparticles makes it possible to perform the same type of measurement with cells as probes and evaluate their own microenvironment. For 18 microm diameter magnetic beads or magnetically labeled stem cells of similar diameter, the viscosity was equivalently equal to 20 mPa s in average. This apparent viscosity was then found to increase with the magnetic probes sizes. The stiffness probed with 100 microm magnetic beads was found in the 50 Pa range, and was lowered by a factor 5 when probed with cells aggregates. The magnetic forces were also successfully applied to the stem cells to enhance the cell seeding process and impose a well defined spatial organization into the scaffold.

Free-flow magnetophoresis provides a fast and efficient means of continuous flow magnetic separation for the detection of biological analytes, due to the wide variety of magnetic particle surface properties available for binding specific targets. Here, we investigate the effect of temperature changes on the deflection behaviour of magnetic particles in a microfluidic magnetophoresis separation chamber. It was found that the extent of deflection was greatly increased at higher temperatures due to decreased solution viscosity and thus reduced resistance against particle motion. This concept was used to improve the resolution of the separation of 2.8 μm and 1 μm diameter magnetic particles. Hence, controlling the temperature of the separation system provides a simple but highly effective means of enhancing magnetic separation efficiency. This concept could also be applied to the temperature-based tuning of microparticle trajectories in many others types of continuous flow processes, such as those using optical, electrical or acoustic forces.

Free-flow magnetophoresis is becoming a popular technique for the separation and manipulation of magnetic particles and materials in microsystems. A wide variety of magnetic particles are commercially available that differ greatly in size and in magnetic properties. To investigate the suitability of different particle types for magnetophoretic operations in microfluidic devices, we compared a range of particles from three manufacturers by pumping them through a microfluidic separation chamber and deflecting them from the direction of laminar flow by applying an external magnetic field. The on-chip deflection of particles was compared to data provided by the manufacturers and magnetisation data obtained from vibrating sample magnetometer (VSM) measurements. Additionally, the extent of deflection was examined over a range of temperatures. Deflection distances were found to increase significantly with increasing temperature. Further to this, a separation of 2.8 and 1.0μm magnetic particles was performed at different temperatures. Separation resolution was found to improve at higher temperatures. Hence, temperature manipulation provides a simple and effective means for improving a magnetic separation or for controlling the angle at which a particle is deflected from its hydrodynamic flow path.