[CVPR2020/PaperSummary]RandLA-Net: Efficient Semantic Segmentation of Large-Scale Point Clouds

Fig1 : In each layer of RandLA-Net, the large-scale point cloud is significantly downsampled, yet is capable of retaining features necessary for accurate segmentation
Fig2 : The proposed local feature aggregation module
Fig3: LocSE module
Fig4: Attentive Pooling module
Eq2: Softmax function of shared MLP
Eq3: Score summation function
Fig5: Dilated Residual Block
Fig6:Illustration of the dilated residual block which significantly increases the receptive field (dotted circle) of each point, colored points represent the aggregated features. L: Local spatial encoding, A: Attentive pooling.
Fig7:The network structure of RandLA-Net. (N, D) represent the number of points and feature dimensions, respectively. FC: fully connected layer, LFA: local feature aggregation, RS: random sampling, MLP: shared multilayer perceptron, US: upsampling, DP: Dropout
Fig8: Time and memory consumption of different sampling approaches. The dashed lines represent estimated values due to the limited GPU memory
Table 1 : The computation time, network parameters and maximum number of input points of different approaches for semantic segmentation on Sequence 08 of SemanticKITTI
Table2 : Quantitative results of different approaches on Semantic3D (reduced-8)
Table 3. Quantitative results of different approaches on SemanticKITTI [7]
Table 4. Quantitative results of different approaches on S3DIS dataset [8] (6-fold cross validation)
Table 5. The mean IoU scores of all ablated networks based on our full RandLA-Net

--

--

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store