English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

In this paper, we propose binary sparse convolutional networks called BSC-Netfor efficient point cloud analysis. We empirically observe that sparseconvolution operation causes larger quantization errors than standardconvolution. However, conventional network quantization methods directlybinarize the weights and activations in sparse convolution, resulting inperformance drop due to the significant quantization loss. On the contrary, wesearch the optimal subset of convolution operation that activates the sparseconvolution at various locations for quantization error alleviation, and theperformance gap between real-valued and binary sparse convolutional networks isclosed without complexity overhead. Specifically, we first present the shiftedsparse convolution that fuses the information in the receptive field for theactive sites that match the pre-defined positions. Then we employ thedifferentiable search strategies to discover the optimal opsitions for activesite matching in the shifted sparse convolution, and the quantization errorsare significantly alleviated for efficient point cloud analysis. For fairevaluation of the proposed method, we empirically select the recently advancesthat are beneficial for sparse convolution network binarization to construct astrong baseline. The experimental results on Scan-Net and NYU Depth v2 showthat our BSC-Net achieves significant improvement upon our srtong baseline andoutperforms the state-of-the-art network binarization methods by a remarkablemargin without additional computation overhead for binarizing sparseconvolutional networks.

Binarizing Sparse Convolutional Networks for Efficient Point Cloud Analysis