Skeleton-bridged Point Completion: From Global Inference to Local Adjustment

Authors: Yinyu Nie, Yiqun Lin, Xiaoguang Han, Shihui Guo, Jian Chang, Shuguang Cui, Jian.J Zhang

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The qualitative and quantitative experiments on both point cloud and mesh completion show that our approach outperforms the existing methods on various object categories.
Researcher Affiliation Academia Yinyu Nie1,2, Yiqun Lin2 Xiaoguang Han2, Shihui Guo3 Jian Chang1 Shuguang Cui2 Jian Jun Zhang1 1Bournemouth University 2SRIBD, CUHKSZ 3Xiamen University
Pseudocode No The paper describes the network architecture and modules in text and figures, but does not include structured pseudocode or algorithm blocks.
Open Source Code No The paper does not include any statement or link indicating that the source code for the methodology is openly available.
Open Datasets Yes Datasets. Two datasets are used for our training. 1) Shape Net-Skeleton [31] for skeleton extraction, and 2) Shape Net Core [4] for surface completion.
Dataset Splits Yes We adopt the train/validation/test split from [39] with five categories (i.e., chair, airplane, rifle, lamp, table) and 15,338 models in total.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper does not specify any software dependencies with version numbers.
Experiment Setup Yes We adopt the batch size at 16 and the learning rate at 1e-3, which decreases by the scale of 0.5 if there is no loss drop within five epochs. 200 epochs are used in total. The weights used in the loss functions are: λk, λs = 1, λm = 0.1, λn = 0.001, λr = 0.1, λG = 0.01.