Selective training of neural networks using motion estimation
Selective training of neural networks using motion estimation, including: selecting, from a plurality of pixels in video data from a vehicle, based on motion relative to the vehicle, one or more pixels; and training a neural network based on the video data by zeroing out an error function applied to the selected one or more pixels.
1. A method comprising: selecting, from a plurality of pixels in video data from a vehicle, based on motion relative to the vehicle, one or more pixels; and
training a neural network based on the video data including the selected one or more pixels by setting the output of an error function applied to the selected one or more pixels during training of the neural network to zero;
wherein selecting the one or more pixels comprises selecting, for inclusion in the one or more pixels, at least one of: one or more pixels corresponding to one or more objects appearing to move away from the autonomous vehicle, or one or more pixels corresponding to one or more objects appearing effectively stationary relative to the autonomous vehicle.
2. The method of claim 1 , wherein selecting the one or more pixels comprises:
identifying, for each pixel of the video data, a corresponding motion vector; and
determining, based on each pixel and the corresponding motion vector, whether to include a respective pixel in the selected one or more pixels.
3. The method of claim 2 , wherein identifying, for each pixel of the video data, the corresponding motion vector comprises providing the video data to another neural network.
4. The method of claim 1 , further comprising providing the trained neural network to one or more autonomous vehicles.
5. The method of claim 1 , wherein the trained neural network is configured to determine, based on sensor data, one or more autonomous vehicle control operations.
6. The method of claim 1 , wherein training the machine learning model comprises providing, to the machine learning model, the video data and one or more control operations of the vehicle corresponding to the video data.
7. An apparatus configured to perform steps comprising:
selecting, from a plurality of pixels in video data from a vehicle, based on motion relative to the vehicle, one or more pixels; and
raining a neural network based on the video data including the selected one or more pixels by setting the output of an error function applied to the selected one or more pixels during training of the neural network to zero;
wherein identifying the one or more pixels comprises identifying, for inclusion in the one or more pixels, at least one of: one or more pixels corresponding to one or more objects appearing to move away from the autonomous vehicle, or one or more pixels corresponding to one or more objects appearing effectively stationary relative to the autonomous vehicle.
8. The apparatus of claim 7 , wherein selecting the one or more pixels comprises:
identifying, for each pixel of the video data, a corresponding motion vector; and
determining, based on each pixel and the corresponding motion vector, whether to include a respective pixel in the selected one or more pixels.
9. The apparatus of claim 8 , wherein identifying, for each pixel of the video data, the corresponding motion vector comprises providing the video data to another neural network.
10. The apparatus of claim 7 , wherein the steps further comprise providing the trained neural network to one or more autonomous vehicles.
11. The apparatus of claim 7 , wherein the trained neural network is configured to determine, based on sensor data, one or more autonomous vehicle control operations.
12. The apparatus of claim 7 , wherein training the machine learning model comprises providing, to the machine learning model, the video data and one or more control operations of the vehicle corresponding to the video data.
13. A computer program product disposed upon a non-transitory computer readable medium, the computer program product comprising computer program instructions that, when executed, cause a computer system to carry out the steps of:
selecting, from a plurality of pixels in video data from a vehicle, based on motion relative to the vehicle, one or more pixels; and
training a neural network based on the video data including the selected one or more pixels by setting the output of an error function applied to the selected one or more pixels during training of the neural network to zero;
wherein selecting the one or more pixels comprises selecting, for inclusion in the one or more pixels, at least one of: one or more pixels corresponding to one or more objects appearing to move away from the autonomous vehicle, or one or more pixels corresponding to one or more objects appearing effectively stationary relative to the autonomous vehicle.
14. The computer program product of claim 13 , wherein selecting the one or more pixels comprises:
identifying, for each pixel of the video data, a corresponding motion vector; and
determining, based on each pixel and the corresponding motion vector, whether to include a respective pixel in the selected one or more pixels.
15. The computer program product of claim 14 , wherein identifying, for each pixel of the video data, the corresponding motion vector comprises providing the video data to another neural network.
16. The computer program product of claim 13 , wherein the steps further comprise providing the trained neural network to one or more autonomous vehicles.
17. The computer program product of claim 13 , wherein the trained neural network is configured to determine, based on sensor data, one or more autonomous vehicle control operations.