Abstract: A method, apparatus, and computer program product are provided for providing improved neural network implementations using a repeated convolution-based attention module. Example embodiments implement a repeated convolution-based attention module that utilizes multiple iterations of a repeated convolutional application layer and subsequent augmentations to generate an attention module output. Example methods may include augmenting an attention input data object based on a previous iteration convolutional output to produce a current iteration input parameter, inputting the input parameter to a repeated convolutional application layer to generate a current iteration input parameter, repeating for multiple iterations, and augmenting the attention input data object based on the final convolutional output to produce an attention module output.

Abstract: Provided herein is a method, apparatus, and computer program product for classifying objects as static objects or dynamic objects based on point cloud data. Methods may include: receiving point cloud data representative of an environment; computing voxel sequences from the point cloud data; extracting voxel-wise semantic features from the voxel sequences; modeling voxel-wise temporal changes based on the voxel-wise semantic features; and classifying objects in the environment as dynamic objects or static objects based on the modeled voxel-wise temporal changes. Computing voxel sequences from the point cloud data may include using a voxel cloud connectivity segmentation method to group voxels in point clouds into perceptually meaningful regions.

Abstract: Segmentation of three dimensional objects may be implemented using a neural network model, a clustering module, a factorization module, and a geometric fitting module. The neural network model is configured to analyze point cloud data for a geographic region and assign probability values outputted from the neural network to points in the point cloud data. The clustering module is configured to group a subset of the probability values based on relative locations of the assigned points in the point cloud data. The factorization module is configured to factor a matrix with the subset of the clustered probability values to assign a line for a three dimensional object of the geographic region. The geometric fitting module is configured to fit at least one predetermined shape for the three dimensional object to the point cloud data based at least on the assigned line.

Abstract: A method, apparatus, and computer program product are provided for providing improved neural network implementations using a repeated convolution-based attention module. Example embodiments implement a repeated convolution-based attention module that utilizes multiple iterations of a repeated convolutional application layer and subsequent augmentations to generate an attention module output. Example methods may include augmenting an attention input data object based on a previous iteration convolutional output to produce a current iteration input parameter, inputting the input parameter to a repeated convolutional application layer to generate a current iteration input parameter, repeating for multiple iterations, and augmenting the attention input data object based on the final convolutional output to produce an attention module output.

Abstract: Provided herein is a method, apparatus, and computer program product for classifying objects as static objects or dynamic objects based on point cloud data. Methods may include: receiving point cloud data representative of an environment; computing voxel sequences from the point cloud data; extracting voxel-wise semantic features from the voxel sequences; modeling voxel-wise temporal changes based on the voxel-wise semantic features; and classifying objects in the environment as dynamic objects or static objects based on the modeled voxel-wise temporal changes. Computing voxel sequences from the point cloud data may include using a voxel cloud connectivity segmentation method to group voxels in point clouds into perceptually meaningful regions.

Abstract: Segmentation of three dimensional objects may be implemented using a neural network model, a clustering module, a factorization module, and a geometric fitting module. The neural network model is configured to analyze point cloud data for a geographic region and assign probability values outputted from the neural network to points in the point cloud data. The clustering module is configured to group a subset of the probability values based on relative locations of the assigned points in the point cloud data. The factorization module is configured to factor a matrix with the subset of the clustered probability values to assign a line for a three dimensional object of the geographic region. The geometric fitting module is configured to fit at least one predetermined shape for the three dimensional object to the point cloud data based at least on the assigned line.