Submitted anonymously.

Submission data

Full name	Long Range Pooling for 3D Large-Scale Scene Understanding
Description	Inspired by the success of recent vision transformers (ViTs) and large kernel design in convolutional neural net- works (CNNs), in this paper, we analyse and explore es- sential reasons for their success and claim two factors that are critical for 3D large-scale scene understanding, i.e., (1) larger receptive field and (2) more non-linear operations. The former is responsible for global context and the lat- ter can enhance the capacity of the network. To achieve above properties, we propose a simple yet effective long range pooling (LRP) module by using dilated max pooling
Input Data Types	Uses Color        Uses 3D
Programming language(s)	python
Hardware	3090
Submission creation date	9 Nov, 2022
Last edited	9 Nov, 2022

3D semantic label results