Bolin Chen

Bolin Chen, Ru-Ling Liao, Jie Chen et al.

Beyond traditional hybrid-based video codec, generative video codec could achieve promising compression performance by evolving high-dimensional signals into compact feature representations for bitstream compactness at the encoder side and developing explicit motion fields as intermediate supervision for high-quality reconstruction at the decoder side. This paradigm has achieved significant success in face video compression. However, compared to facial videos, human body videos pose greater challenges due to their more complex and diverse motion patterns, i.e., when using explicit motion guidance for Generative Human Video Coding (GHVC), the reconstruction results could suffer severe distortions and inaccurate motion. As such, this paper highlights the limitations of explicit motion-based approaches for human body video compression and investigates the GHVC performance improvement with the aid of Implicit Motion Transformation, namely IMT. In particular, we propose to characterize complex human body signal into compact visual features and transform these features into implicit motion guidance for signal reconstruction. Experimental results demonstrate the effectiveness of the proposed IMT paradigm, which can facilitate GHVC to achieve high-efficiency compression and high-fidelity synthesis.

Bolin Chen, Dex Doksoo Lee, Wei "Wayne'' Chen et al.

Metamaterials design for advanced functionality often entails the inverse design on nonlinear and condition-dependent responses (e.g., stress-strain relation and dispersion relation), which are described by continuous functions. Most existing design methods focus on vector-valued responses (e.g., Young's modulus and bandgap width), while the inverse design of functional responses remains challenging due to their high-dimensionality, the complexity of accommodating design requirements in inverse-design frameworks, and non-existence or non-uniqueness of feasible solutions. Although generative design approaches have shown promise, they are often data-hungry, handle design requirements heuristically, and may generate infeasible designs without uncertainty quantification. To address these challenges, we introduce a RAndom-forest-based Generative approach (RAG). By leveraging the small-data compatibility of random forests, RAG enables data-efficient predictions of high-dimensional functional responses. During the inverse design, the framework estimates the likelihood through the ensemble which quantifies the trustworthiness of generated designs while reflecting the relative difficulty across different requirements. The one-to-many mapping is addressed through single-shot design generation by sampling from the conditional likelihood. We demonstrate RAG on: 1) acoustic metamaterials with prescribed partial passbands/stopbands, and 2) mechanical metamaterials with targeted snap-through responses, using 500 and 1057 samples, respectively. Its data-efficiency is benchmarked against neural networks on a public mechanical metamaterial dataset with nonlinear stress-strain relations. Our framework provides a lightweight, trustworthy pathway to inverse design involving functional responses, expensive simulations, and complex design requirements, beyond metamaterials.

Shanzhi Yin, Bolin Chen, Xinju Wu et al.

This paper proposes an efficient 3D avatar coding framework that leverages compact human priors and canonical-to-target transformation to enable high-quality 3D human avatar video compression at ultra-low bit rates. The framework begins by training a canonical Gaussian avatar using articulated splatting in a network-free manner, which serves as the foundation for avatar appearance modeling. Simultaneously, a human-prior template is employed to capture temporal body movements through compact parametric representations. This decomposition of appearance and temporal evolution minimizes redundancy, enabling efficient compression: the canonical avatar is shared across the sequence, requiring compression only once, while the temporal parameters, consisting of just 94 parameters per frame, are transmitted with minimal bit-rate. For each frame, the target human avatar is generated by deforming canonical avatar via Linear Blend Skinning transformation, facilitating temporal coherent video reconstruction and novel view synthesis. Experimental results demonstrate that the proposed method significantly outperforms conventional 2D/3D codecs and existing learnable dynamic 3D Gaussian splatting compression method in terms of rate-distortion performance on mainstream multi-view human video datasets, paving the way for seamless immersive multimedia experiences in meta-verse applications.