Selected work

Representative Papers

EigenGS Representation: From Eigenspace to Gaussian Image Space

Mar 10, 2025

This work addresses the slow initialization and difficulty in multi-scale modeling—leading to high-frequency artifacts—in 2D Gaussian splatting reconstruction. We propose a PCA-guided, frequency-aware Gaussian parameterization method. By establishing an end-to-end differentiable mapping from a learned feature subspace to Gaussian ellipsoid parameters (center, covariance, opacity), our approach enables instantaneous initialization of Gaussian parameters for novel views. A frequency-aware learning mechanism further allows ellipsoids to adaptively capture multi-scale spatial structures. To the best of our knowledge, this is the first work to establish a generalizable, differentiable mapping paradigm between feature space and Gaussian image representations. Evaluated on multi-resolution and multi-category datasets, our method achieves superior reconstruction quality over direct 2D Gaussian fitting, reduces parameter count by 37%, accelerates training by 5.2×, and supports real-time, high-fidelity image representation.

0 citationsRead paper

Segment Anything, Even Occluded

Mar 08, 2025

This work addresses amodal instance segmentation—simultaneously detecting and segmenting both visible and occluded parts of objects—to improve perception robustness under occlusion. Existing methods suffer from inflexible joint training of detectors and mask decoders, hindering reuse of pre-trained detectors. To overcome this, we propose SAMEO: the first framework that adapts Segment Anything Model (SAM) as a general-purpose, plug-and-play mask decoder compatible with arbitrary front-end detectors. We further introduce Amodal-LVIS, the first large-scale synthetic dataset (300K images) for amodal segmentation, alleviating the scarcity of real-world occlusion annotations. Additionally, we design a zero-shot cross-domain transfer strategy leveraging synthetic data. On COCOA-cls and D2SA benchmarks, SAMEO achieves state-of-the-art zero-shot performance, demonstrating significantly improved generalization to unseen occlusion patterns without fine-tuning.

0 citationsRead paper

Recent publications

Latest Papers

EigenGS Representation: From Eigenspace to Gaussian Image Space

Mar 10, 2025

This work addresses the slow initialization and difficulty in multi-scale modeling—leading to high-frequency artifacts—in 2D Gaussian splatting reconstruction. We propose a PCA-guided, frequency-aware Gaussian parameterization method. By establishing an end-to-end differentiable mapping from a learned feature subspace to Gaussian ellipsoid parameters (center, covariance, opacity), our approach enables instantaneous initialization of Gaussian parameters for novel views. A frequency-aware learning mechanism further allows ellipsoids to adaptively capture multi-scale spatial structures. To the best of our knowledge, this is the first work to establish a generalizable, differentiable mapping paradigm between feature space and Gaussian image representations. Evaluated on multi-resolution and multi-category datasets, our method achieves superior reconstruction quality over direct 2D Gaussian fitting, reduces parameter count by 37%, accelerates training by 5.2×, and supports real-time, high-fidelity image representation.

0 citationsRead paper

Segment Anything, Even Occluded

Mar 08, 2025

This work addresses amodal instance segmentation—simultaneously detecting and segmenting both visible and occluded parts of objects—to improve perception robustness under occlusion. Existing methods suffer from inflexible joint training of detectors and mask decoders, hindering reuse of pre-trained detectors. To overcome this, we propose SAMEO: the first framework that adapts Segment Anything Model (SAM) as a general-purpose, plug-and-play mask decoder compatible with arbitrary front-end detectors. We further introduce Amodal-LVIS, the first large-scale synthetic dataset (300K images) for amodal segmentation, alleviating the scarcity of real-world occlusion annotations. Additionally, we design a zero-shot cross-domain transfer strategy leveraging synthetic data. On COCOA-cls and D2SA benchmarks, SAMEO achieves state-of-the-art zero-shot performance, demonstrating significantly improved generalization to unseen occlusion patterns without fine-tuning.

0 citationsRead paper