Abstract: Techniques are disclosed for training a feature extraction model. A computing device can receive a training image and generate noised feature vectors using a feature extraction model characterized by first parameters and taking the training image as input. The computing device can determine the noised feature vectors by at least determining a feature vector for individual pixels in the training image and applying noise to each feature vector. The computing device can generate a reconstructed image using a reconstructor model characterized by second parameters and taking the noised feature vectors as input. The computing device can determine a reconstruction loss by comparing the training image with the reconstructed image and a noise loss using the noise applied to each feature vector. The computing device can update the first parameters based on the noise loss.

Abstract: Systems and methods for moving point detection for point clouds. Some implementations may include obtaining a sequence of point clouds, wherein the sequence of point clouds includes a current point cloud, a previous point cloud that precedes the current point cloud, and a next point cloud that follows the current point cloud; inputting a voxelized representation of the current point cloud to a forward-pass recurrent neural network with a hidden state based on the previous point cloud to obtain a first voxelized activation; inputting the voxelized representation of the current point cloud to a backward-pass recurrent neural network with a hidden state based on the next point cloud to obtain a second voxelized activation; combining the first voxelized activation and the second voxelized activation to obtain a prediction of whether respective points are moving or static; and updating the current point cloud based on the prediction.

Abstract: Systems and methods for semantic labeling of point clouds using images. Some implementations may include obtaining a point cloud that is based on lidar data reflecting one or more objects in a space; obtaining an image that includes a view of at least one of the one or more objects in the space; determining a projection of points from the point cloud onto the image; generating, using the projection, an augmented image that includes one or more channels of data from the point cloud and one or more channels of data from the image; inputting the augmented image to a two dimensional convolutional neural network to obtain a semantic labeled image wherein elements of the semantic labeled image include respective predictions; and mapping, by reversing the projection, predictions of the semantic labeled image to respective points of the point cloud to obtain a semantic labeled point cloud.

Abstract: Systems and methods for semantic labeling of point clouds using images. Some implementations may include obtaining a point cloud that is based on lidar data reflecting one or more objects in a space; obtaining an image that includes a view of at least one of the one or more objects in the space; determining a projection of points from the point cloud onto the image; generating, using the projection, an augmented image that includes one or more channels of data from the point cloud and one or more channels of data from the image; inputting the augmented image to a two dimensional convolutional neural network to obtain a semantic labeled image wherein elements of the semantic labeled image include respective predictions; and mapping, by reversing the projection, predictions of the semantic labeled image to respective points of the point cloud to obtain a semantic labeled point cloud.

Abstract: Systems and methods for moving point detection for point clouds. Some implementations may include obtaining a sequence of point clouds, wherein the sequence of point clouds includes a current point cloud, a previous point cloud that precedes the current point cloud, and a next point cloud that follows the current point cloud; inputting a voxelized representation of the current point cloud to a forward-pass recurrent neural network with a hidden state based on the previous point cloud to obtain a first voxelized activation; inputting the voxelized representation of the current point cloud to a backward-pass recurrent neural network with a hidden state based on the next point cloud to obtain a second voxelized activation; combining the first voxelized activation and the second voxelized activation to obtain a prediction of whether respective points are moving or static; and updating the current point cloud based on the prediction.

Abstract: Systems and methods for semantic labeling of point clouds using images. Some implementations may include obtaining a point cloud that is based on lidar data reflecting one or more objects in a space; obtaining an image that includes a view of at least one of the one or more objects in the space; determining a projection of points from the point cloud onto the image; generating, using the projection, an augmented image that includes one or more channels of data from the point cloud and one or more channels of data from the image; inputting the augmented image to a two dimensional convolutional neural network to obtain a semantic labeled image wherein elements of the semantic labeled image include respective predictions; and mapping, by reversing the projection, predictions of the semantic labeled image to respective points of the point cloud to obtain a semantic labeled point cloud.