Diffusion transformers (DiT) have demonstrated exceptional performance in video generation. However, their large number of parameters and high computational complexity limit their deployment on edge devices. Quantization can reduce storage requirements and accelerate inference by lowering the bit-width of model parameters. Yet, existing quantization methods for image generation models do not generalize well to video generation tasks. We identify two primary challenges: the loss of information during quantization and the misalignment between optimization objectives and the unique requirements of video generation. To address these challenges, we present **Q-VDiT**, a quantization framework specifically designed for video DiT models. From the quantization perspective, we propose the *Token aware Quantization Estimator* (TQE), which compensates for quantization errors in both the token and feature dimensions. From the optimization perspective, we introduce *Temporal Maintenance Distillation* (TMD), which preserves the spatiotemporal correlations between frames and enables the optimization of each frame with respect to the overall video context. Our W3A6 Q-VDiT achieves a scene consistency score of 23.40, setting a new benchmark and outperforming the current state-of-the-art quantization methods by **1.9$\times$**.

3D few-shot class incremental learning (FSCIL) aims to learn new point cloud categories from limited samples while preventing the forgetting of previously learned categories. This research area significantly enhances the capabilities of self-driving vehicles and computer vision systems. Existing 3D FSCIL approaches primarily utilize multimodal pre-trained models to extract the semantic features, heavily dependent on meticulously designed high-quality prompts and fine-tuning strategies. To reduce this dependence, this paper proposes a novel method for **3D** **F**SCI**L** with **E**mbedded **G**eometric features (**3D-FLEG**). Specifically, 3D-FLEG develops a point cloud *geometric feature extraction module* to capture category-related geometric characteristics. To address the modality heterogeneity issues that arise from integrating geometric and text features, 3D-FLEG introduces a *geometric feature embedding module*. By augmenting text prompts with spatial geometric features through these modules, 3D-FLEG can learn robust representations of new categories even with limited samples, while mitigating forgetting of the previously learned categories. Experiments conducted on several publicly available 3D point cloud datasets, including ModelNet, ShapeNet, ScanObjectNN, and CO3D, demonstrate 3D-FLEG's superiority over existing state-of-the-art 3D FSCIL methods. Code is available at https://github.com/lixiangqi707/3D-FLEG.

Co-speech gestures enhance both human-human and human-robot interactions. This paper examines the efficacy of a data-driven approach for generating synchronised co-speech gestures in three social robots to improve social interactions. Building on a sequence-to-sequence model, which maps speech to gestures [21], this work uses the Talking With Hands 16.2M dataset [11] to generate natural gestures for face-to-face conversations. Additionally, we address synchronisation issues identified in the original study. The model’s generality is tested on three robots—NAO, Pepper, and ARI. Objective and subjective evaluations, confirm that a data-driven approach effectively generates synchronised co-speech gestures.

As the resolution continues to increase, the scale of the power grid of the flat panel display (FPD) becomes huge. This imposes a severe computational challenge for analysis and storage. In this paper, based on the highly periodic FPD structure, we present an efficient simulator which contains a novel stamping scheme and a fast structural solver. Experiments on real industrial cases show that compared to the conventional stamping scheme, our proposed stamping scheme achieves up to 118× speedup in 74× less memory; compared to the state-of-art direct solver, Cholmod, our proposed structural solver achieves up to 17× speedup in 5× less memory.

Extensible languages enable the convenient construction of many kinds of domain-specific languages (DSLs) by mapping domain-specific surface syntax into the host language's core forms in a layered and composable way. The host language's debugger, however, reports evaluation and data details in ways that reflect the host language, instead of the DSL in its own terms, and closing the gap may require more than correlating host evaluation steps to the original DSL source. In this paper, we describe an approach to DSL construction with macros that pairs the mapping of DSL terms to host terms with a mapping to convert primitive events back to domain-specific concepts. Domain-specific events are then suitable for presenting to a user or wiring into a domain-specific visualization. We present a core model of evaluation and events, and we present a language design---analogous to pattern-based notations for macros, but in the other direction---for describing how events in a DSL's expansion are mapped to events at the DSL's level.