Search for: All records

Annotating camera poses on dynamic Internet videos at scale is critical for advancing fields like realistic video generation and simulation. However, collecting such a dataset is difficult, as most Internet videos are unsuitable for pose estimation. Furthermore, annotating dynamic Internet videos present significant challenges even for state-of-the-art methods. In this paper, we introduce DynPose-100K, a large-scale dataset of dynamic Internet videos annotated with camera poses. Our collection pipeline addresses filtering using a carefully combined set of task-specific and generalist models. For pose estimation, we combine the latest techniques of point tracking, dynamic masking, and structure-from-motion to achieve improvements over the state-of-the-art approaches. Our analysis and experiments demonstrate that DynPose-100K is both large-scale and diverse across several key attributes, opening up avenues for advancements in various downstream applications 

To fully capitalize on the unique properties of 2D materials, cost-effective techniques for producing high-quality 2D flakes at scale are crucial. In this work, we show that dry ball-milling, a commonly used powder-processing technique, can be effectively and efficiently upgraded into an automated exfoliation technique. It is done by adding polymer as adhesives into a ball mill to mimic the well-known tape exfoliation process, which is known to produce 2D flakes with the highest quality but is limited by its extremely low efficiency on large-scale production. Seventeen types of commonly seen polymers, including both artificial and natural ones, have been examined as additives to dry ball-mill hexagonal boron nitride. A parallel comparison between different additives identifies low-cost natural polymers such as starch as promising dry ball-mill additives to produce ultrathin flakes with the largest aspect ratio. The mechanical, thermal, and surface properties of the polymers are proposed as key features that simultaneously determine the exfoliation efficiency, and their ranking of importance in the mechanical exfoliation process is revealed using a machine learning model. Finally, the potential of the polymer-assisted ball-mill exfoliation method as a universal way to produce ultra-thin 2D nanosheets is also demonstrated. 

Abstract Light triggers an enhancement of global translation during photomorphogenesis in Arabidopsis, but little is known about the underlying mechanisms. The phosphorylation of the α-subunit of eukaryotic initiation factor 2 (eIF2α) at a conserved serine residue in the N-terminus has been shown as an important mechanism for the regulation of protein synthesis in mammalian and yeast cells. However, whether the phosphorylation of this residue in plant eIF2α plays a role in regulation of translation remains elusive. Here, we show that the quadruple mutant of SUPPRESSOR OF PHYA-105 family members (SPA1-SPA4) display repressed translation efficiency after light illumination. Moreover, SPA1 directly phosphorylates the eIF2α C-terminus under light conditions. The C-term-phosphorylated eIF2α promotes translation efficiency and photomorphogenesis, whereas the C-term-unphosphorylated eIF2α results in a decreased translation efficiency. We also demonstrate that the phosphorylated eIF2α enhances ternary complex assembly by promoting its affinity to eIF2β and eIF2γ. This study reveals a unique mechanism by which light promotes translation via SPA1-mediated phosphorylation of the C-terminus of eIF2α in plants. 

As the feature size of microelectronic circuits is scaling down to nanometer order, the increasing interconnect crosstalk, resistance-capacitance (RC) delay and power consumption can limit the chip performance and reliability. To address these challenges, new low-kdielectric (k < 2) materials need to be developed to replace current silicon dioxide (k = 3.9) or SiCOH, etc. However, existing low-kdielectric materials, such as organosilicate glass or polymeric dielectrics, suffer from poor thermal and mechanical properties. Two-dimensional polymers (2DPs) are considered promising low-kdielectric materials because of their good thermal and mechanical properties, high porosity and designability. Here, we report a chemical-vapor-deposition (CVD) method for growing fluoride rich 2DP-F films on arbitrary substrates. We show that the grown 2DP-F thin films exhibit ultra-low dielectric constant (in plane k = 1.85 and out-of-plane k = 1.82) and remarkable mechanical properties (Young’s modulus > 15 GPa). We also demonstrated the improved performance of monolayer MoS₂field-effect-transistors when utilizing 2DP-F thin films as dielectric substrates. 

A deep neural network using rectified linear units represents a continuous piecewise linear (CPWL) function and vice versa. Recent results in the literature estimated that the number of neurons needed to exactly represent any CPWL function grows exponentially with the number of pieces or exponentially in terms of the factorial of the number of distinct linear components. Moreover, such growth is amplified linearly with the input dimension. These existing results seem to indicate that the cost of representing a CPWL function is expensive. In this paper, we propose much tighter bounds and establish a polynomial time algorithm to find a network satisfying these bounds for any given CPWL function. We prove that the number of hidden neurons required to exactly represent any CPWL function is at most a quadratic function of the number of pieces. In contrast to all previous results, this upper bound is invariant to the input dimension. Besides the number of pieces, we also study the number of distinct linear components in CPWL functions. When such a number is also given, we prove that the quadratic complexity turns into bilinear, which implies a lower neural complexity because the number of distinct linear components is always not greater than the minimum number of pieces in a CPWL function. When the number of pieces is unknown, we prove that, in terms of the number of distinct linear components, the neural complexities of any CPWL function are at most polynomial growth for low-dimensional inputs and factorial growth for the worst-case scenario, which are significantly better than existing results in the literature.