Abstract: Disclosed herein are system, method, and computer program product aspects for enabling an autonomous vehicle (AV) to detect objects and forecast their predicted positions. The system can monitor an object within a vicinity of the AV. A plurality of trajectories predicting paths the object will take at a future time can be generated, the plurality of trajectories being based on a generated three-dimensional (3D) point cloud map indicating current and past characteristics of the object. Using a learned model, a forecasted position of the object at an instance in time can be generated along one or more of the plurality of trajectories. A maneuver for the AV can be performed based on the forecasted position.

Abstract: Methods and systems for detecting objects near an autonomous vehicle (AV) are disclosed. An AV will capture an image. A trained network will process the image at a lower resolution and generate a first feature map that classifies object(s) within the image. The system will crop the image and use the network to process the cropped section at a higher resolution to generate a second feature map that classifies object(s) that appear within the cropped section. The system will crop the first feature map to match a corresponding region of the cropped section of the image. The system will fuse the cropped first and second feature maps to generate a third feature map. The system may output the object classifications in the third feature map to an AV system, such as a motion planning system that will use the object classifications to plan a trajectory for the AV.

Abstract: Systems and methods for processing high resolution images are disclosed. The methods include generating a saliency map of a received high-resolution image using a saliency model. The saliency map includes a saliency value associated with each of a plurality of pixels of the high-resolution image. The method then includes using the saliency map for generating an inverse transformation function that is representative of an inverse mapping of one or more first pixel coordinates in a warped image to one or more second pixel coordinates in the high-resolution image, and implementing an image warp for converting the high-resolution image to the warped image using the inverse transformation function. The warped image is a foveated image that includes at least one region having a higher resolution than one or more other regions of the warped image.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for object detection are disclosed. Methods can include, for each of a plurality of locations in one or more positive images, image filters are identified, each image filter representing visual features of a location in a positive image (e.g., an image that includes a particular object). Positive location feature scores and negative location feature scores are determined for locations within images. A positive location feature score is based on a similarity between the image filter and feature values for a positive image. A negative location feature score is determined based on a similarity between the image filter and feature values for a negative image. A distinctive location is identified based on the positive and negative location feature scores, and distinguishing feature values for identifying the particular object are identified for the distinctive location.