PulseAugur
EN
LIVE 18:53:10

New AI methods tackle long-horizon 3D reconstruction challenges

By PulseAugur Editorial · [4 sources] ·

Researchers have developed HorizonStream, a novel Transformer architecture designed to improve long-horizon 3D reconstruction from streaming video. This method addresses memory and time complexity issues by modeling geometric propagation through an evidence influence kernel. HorizonStream achieves state-of-the-art performance by generalizing from short training clips to sequences exceeding 10,000 frames with constant memory and linear time complexity. Separately, the GHOST framework offers a training-free approach to manage the key-value cache in streaming 3D reconstruction, reducing cache size by nearly half and speeding up inference. AI

IMPACT These advancements in streaming 3D reconstruction could enable more robust and efficient real-time applications in robotics, augmented reality, and autonomous systems.

RANK_REASON Multiple papers introducing new methods for 3D reconstruction.

Read on Hugging Face Daily Papers →

AI-generated summary · Google Gemini · from 4 sources. How we write summaries →

New AI methods tackle long-horizon 3D reconstruction challenges

COVERAGE [4]

  1. Hugging Face Daily Papers TIER_1 English(EN) ·

    HorizonStream: Long-Horizon Attention for Streaming 3D Reconstruction

    HorizonStream addresses long-term 3D reconstruction challenges by modeling geometric propagation through an evidence influence kernel, enabling stable, scalable streaming reconstruction with constant memory and linear time complexity.

  2. arXiv cs.CV TIER_1 English(EN) · Leyang Chen, Junyi Wu, Zhiteng Li, Yulun Zhang ·

    GHOST: Geometry-Hierarchical Online Streaming Token Eviction for Efficient 3D Reconstruction

    arXiv:2605.15852v2 Announce Type: replace Abstract: Streaming 3D reconstruction from long monocular video sequences requires maintaining a key-value (KV) cache that grows linearly with sequence length, creating a severe memory bottleneck. Existing approaches either truncate the c…

  3. arXiv cs.CV TIER_1 English(EN) · Chong Cheng, Peilin Tao, Nanjie Yao, Guanzhi Ding, Xianda Chen, Yuansen Du, Xiaoyang Guo, Wei Yin, Weiqiang Ren, Qian Zhang, Zhengqing Chen, Hao Wang ·

    HorizonStream: Long-Horizon Attention for Streaming 3D Reconstruction

    arXiv:2605.23889v1 Announce Type: new Abstract: Online 3D reconstruction requires estimating camera pose and scene geometry under strict causal and bounded-memory constraints. Existing methods often suffer from drift, jitter, or collapse on long sequences. We trace these failures…

  4. arXiv cs.CV TIER_1 English(EN) · Hao Wang ·

    HorizonStream: Long-Horizon Attention for Streaming 3D Reconstruction

    Online 3D reconstruction requires estimating camera pose and scene geometry under strict causal and bounded-memory constraints. Existing methods often suffer from drift, jitter, or collapse on long sequences. We trace these failures to a fundamental mismatch. Streaming geometry i…

RELATED ENTITIES

RELATED TOPICS