Multi-tier network for task-oriented deep neural network
A method for controlling an autonomous vehicle is disclosed. The method may include receiving image data. The image data may be logistically classified by a first neural network as pertaining to one situation of a plurality of situations. Based on this logistic classification, the image data may be assigned to a second neural network that is specifically trained to address the one situation. The second neural network may perform regression on the image data. Thereafter, the vehicle may be control with a command based on the regression.
1. A method comprising:
running a computer system comprising (1) a supervisory neural network trained to perform classification with respect to a plurality of classes and (2) a plurality of specialized neural networks each trained to perform regression with respect to a different class of the plurality of classes;
receiving, by the computer system, image data;
classifying, by the supervisory neural network, the image data as pertaining to a first class of the plurality of classes;
assigning, by the computer system as a result of the classifying, the image data to a first specialized neural network of the plurality of specialized neural networks, wherein the first specialized neural network is specifically trained to perform regression with respect to the first class;
performing, by the first specialized neural network, regression on the image data; and
implementing, by the computer system, a command based on the regression.
2. The method of claim 1 , wherein the computer system is carried onboard a vehicle.
3. The method of claim 2 , wherein the command is a steering command for the vehicle.
4. The method of claim 3 , wherein the plurality of classes comprises a drive straight class, a turn left class, and a turn right class.
5. The method of claim 4 , wherein the first specialized neural network is trained to perform regression on images that have been classified as pertaining to the drive straight class and the plurality of specialized neural networks further comprises:
a second specialized neural network specifically trained to perform regression on images that have been classified as pertaining to the turn left class; and
a third specialized neural network specifically trained to perform regression on images that have been classified as pertaining to the turn right class.
6. The method of claim 4 , wherein the first class is selected from the group consisting of the drive straight class, the turn left class, and the turn right class.
7. The method of claim 1 , further comprising:
receiving, by the computer system, other image data;
classifying, by the supervisory neural network, the other image data as pertaining to a second class of the plurality of classes;
assigning, by the computer system as a result of the classifying of the other image data, the other image data to a second specialized neural network of the plurality of specialized neural networks, wherein the second specialized neural network is specifically trained to perform regression with respect to the second class;
performing, by the second specialized neural network, regression on the other image data; and
implementing, by the computer system, a subsequent command based on the regression on the other image data.
8. A multi-tiered method of processing image data in order to perform autonomous driving of a vehicle, the method comprising:
running a computer system onboard a vehicle, the computer system comprising (1) a supervisory neural network trained to perform classification with respect to a plurality of classes and (2) a plurality of specialized neural networks each trained to perform regression with respect to a different class of the plurality of classes;
capturing, by a camera carried onboard the vehicle, at least one image of an area in front of the vehicle;
classifying, by the supervisory neural network, the at least one image as pertaining to a first class of the plurality of classes;
assigning, by the computer system as a result of the classifying, the at least one image to a first specialized neural network of the plurality of specialized neural networks, wherein the first specialized neural network is specifically trained to perform regression with respect to the first class;
performing, by the first specialized neural network, regression on the at least one image; and
implementing, by the computer system, a command based on the regression, the command affecting the driving of the vehicle.
9. The method of claim 8 , wherein the command is a steering command for the vehicle.
10. The method of claim 9 , wherein the plurality of classes comprises a drive straight class, a turn left class, and a turn right class.
11. The method of claim 10 , wherein the first specialized neural network is trained to perform regression on images that have been classified as pertaining to the drive straight class.
12. The method of claim 11 , wherein the plurality of specialized neural networks further comprises:
a second specialized neural network specifically trained to perform regression on images that have been classified as pertaining to the turn left class; and
a third specialized neural network specifically trained to perform regression on images that have been classified as pertaining to the turn right class.
13. A system comprising:
at least one processor carried on-board an autonomous vehicle; and
first memory operably connected to the at least one processor, the first memory storing software programmed to
provide a supervisory neural network trained to classify images as pertaining to one class of a plurality of classes,
provide a plurality of specialized neural networks, wherein each specialized neural network of the plurality of specialized neural networks is specifically trained to perform regression on images pertaining to a different class of the plurality of classes,
receive image data,
feed the image data to the supervisory neural network,
feed the image data to a specific specialized neural network of the plurality of specialized neural networks based on the classification performed by the supervisory neural network on the image data, and
control operation of the autonomous vehicle in accordance with the regression performed by the specific specialized neural network on the image data.
14. The system of claim 13 , wherein the first memory stores software programmed to steer the autonomous vehicle in accordance with the regression performed by the specific specialized neural network on the image data.
15. The system of claim 14 , wherein the plurality of classes comprises a drive straight class, a turn left class, and a turn right class.
16. The system of claim 15 , wherein the plurality of specialized neural networks comprises:
a first specialized neural network specifically trained to perform regression on images that have been classified as pertaining to the drive straight class;
a second specialized neural network specifically trained to perform regression on images that have been classified as pertaining to the turn left class; and
a third specialized neural network specifically trained to perform regression on images that have been classified as pertaining to the turn right class.