Detecting, identifying, and tracking objects based upon a tensor with sparsely computed non-zero values

A method performed by a computing system for generating an overhead image of a scene based upon a perspective image of the scene, the method comprising:

• obtaining the perspective image from a camera;
• generating a feature image based upon the perspective image, where the feature image comprises X×Y pixels, and further where each pixel in the X×Y pixels has an N-dimensional feature vector that includes N feature values assigned thereto;
• providing the feature image as input to a machine learning model, where the machine learning model generates a tensor based upon the feature image, where the tensor comprises, for each pixel in the X×Y pixels, several depth bins that include values that are representative of likelihoods that an object exists in the scene at distances from the camera that correspond to the depth bins, and further where the machine learning model is trained using a Gumbel-softmax algorithm;
• and
• generating the image in the overhead view based upon the tensor output by the machine learning model.

A computer-readable storage medium comprising instructions that, when executed by a processor, cause the processor to perform acts comprising:

• obtaining a perspective image from a camera;
• generating a feature image based upon the perspective image, where the feature image comprises X×Y pixels, and further where each pixel in the X×Y pixels has an N-dimensional feature vector that includes N feature values assigned thereto;
• generating, by a machine learning model, a tensor based upon the feature image, where the tensor comprises, for each pixel in the X×Y pixels, several bins that include values that are representative of likelihoods that an object exists in the scene at distances from the camera that correspond to the depth bins, and further where the machine learning model is trained using a Gumbel-softmax algorithm;
• and
• generating an image in an overhead view based upon the tensor generated by the machine learning model.