Eurographics Partner Events

Permanent URI for this community

https://diglib.eg.org/handle/10.2312/974

Browse

Browsing Eurographics Partner Events by Subject "3D imaging"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    Label Name is Mantra: Unifying Point Cloud Segmentation across Heterogeneous Datasets
    (The Eurographics Association, 2024) Liang, Yixun; He, Hao; Xiao, Shishi; Lu, Hao; Chen, Yingcong; Chen, Renjie; Ritschel, Tobias; Whiting, Emily
    Point cloud segmentation is a fundamental task in 3D vision that serves a wide range of applications. Despite recent advancements, its practical usability is still limited by the availability of training data. The prevalent methodologies cannot optimally exploit multiple datasets due to the inconsistency of labels across datasets. In this work, we introduce a robust method that accommodates learning from diverse datasets with variant label sets. We leverage a pre-trained language model to map discrete labels into a continuous latent space using their semantic names. This harmonizes labels across datasets, facilitating concurrent training. Contrarily, when classifying points within the continuous 3D space via their linguistic tokens, our model exhibits superior generalizability compared to extant methods with fixed decoder structures. Further, our approach assimilates prompt learning to alleviate data shifts across sources. Comprehensive evaluations attest that our model markedly surpasses current benchmarks.
  • Thumbnail Image
    Item
    SuBloNet: Sparse Super Block Networks for Large Scale Volumetric Fusion
    (The Eurographics Association, 2021) Rückert, Darius; Stamminger, Marc; Andres, Bjoern and Campen, Marcel and Sedlmair, Michael
    Training and inference of convolutional neural networks (CNNs) on truncated signed distance fields (TSDFs) is a challenging task. Large parts of the scene are usually empty, which makes dense implementations inefficient in terms of memory consumption and compute throughput. However, due to the truncation distance, non-zero values are grouped around the surface creating small dense blocks inside the large empty space. We show that this structure can be exploited by storing the TSDF in a block sparse tensor and then decomposing it into rectilinear super blocks. A super block is a dense 3d cuboid of variable size and can be processed by conventional CNNs. We analyze the rectilinear decomposition and present a formulation for computing the bandwidth-optimal solution given a specific network architecture. However, this solution is NP-complete, therefore we also a present a heuristic approach for fast training and inference tasks. We verify the effectiveness of SuBloNet and report a speedup of 4x towards dense implementations and 1.7x towards state-of-the-art sparse implementations. Using the super block architecture, we show that recurrent volumetric fusion is now possible on large scale scenes. Such a systems is able to reconstruct high-quality surfaces from few noisy depth images.