Using decay parameters for inferencing with neural networks
Apparatuses, systems, and techniques to identify objects with in an image. In at least one embodiment, objects are identified in an image using one or more neural networks, in which the one or more neural networks are trained using one or more decay parameters.
1. One or more processors, comprising:
circuitry to use one or more neural networks to determine a location of one or more unlabeled versions of one or more objects depicted within two or more first frames of a video based, at least in part, on motion of one or more labeled versions of the one or more objects depicted within one or more second frames of the video, wherein the two or more first frames comprise a larger subset of frames of the video than the one or more second frames.
2. The one or more processors of claim 1 , wherein the circuitry is further configured to:
apply one or more decay parameters to prior state information maintained for the one or more neural networks and used for identifying the one or more objects, a weighting of the prior state information being reduced according to the one or more decay parameters.
3. The one or more processors of claim 2 , wherein the circuitry is further configured to:
store the prior state information external to the one or more neural networks and provide the prior state information to the one or more neural networks for each set of input to the one or more neural networks.
4. The one or more processors processer of claim 2 , wherein the circuitry is further configured to:
obtain a continuous stream of video data; and
provide individual frames of the video data to the one or more neural networks for use in identifying one or more digital representations and updating the prior state information.
5. The one or more processors of claim 1 , wherein the circuitry is further configured to:
determine one or more decay parameters using a hyper-optimization process and a selected decay function.
6. The one or more processors of claim 1 , wherein the one or more neural networks are trained using sequences of sparse video data, individual sequences including at least one labeled version of the one or more objects depicted within the one or more second frames of the video, and wherein the one or more second frames of the video are used to simulate state information and the two or more first frames of the video are used to update the state information and identify one or more digital representations of the one or more objects represented in the individual sequence.
7. The one or more processors of claim 1 , wherein the one or more neural networks include at least one recurrent neural network (RNN) and one or more decay parameters are determined according to an exponential decay function.
8. A system comprising:
one or more processors to be configured to use one or more neural networks to determine a location of one or more unlabeled versions of one or more objects depicted within two or more first frames of a video based, at least in part, on motion of one or more labeled versions of the one or more objects depicted within one or more second frames of the video, wherein the two or more first frames comprise a larger subset of frames of the video than the one or more second frames; and
one or more memories to store the one or more neural networks for determining the location.
9. The system of claim 8 , wherein the one or more processors are further to be configured to:
apply one or more decay parameters to prior state information maintained for the one or more neural networks and used for identifying one or more digital representations, a weighting of the prior state information being reduced according to the one or more decay parameters.
10. The system of claim 9 , wherein the one or more processors are further to be configured to:
store the prior state information external to the one or more neural networks and provide the prior state information to the one or more neural networks for each set of input to the one or more neural networks.
11. The system of claim 9 , wherein the one or more processors are further to be configured to:
obtain a continuous stream of video data; and
provide individual frames of the video data to the one or more neural networks for use in identifying the one or more digital representations and updating the prior state information.
12. The system of claim 8 , wherein the one or more processors are further to be configured to:
determine one or more decay parameters using a hyper-optimization process and a selected decay function.
13. The system of claim 8 , wherein the one or more neural networks are trained using sequences of sparse video data, individual sequences including at least one labeled version of the one or more objects depicted within one of the one or more second frames of the video, and wherein the one or more second frames of the video are used to simulate state information and the two or more first frames of the video are used to update the state information and identify one or more digital representations of the one or more objects represented in the individual sequence.
14. The system of claim 8 , wherein the one or more neural networks include at least one recurrent neural network (RNN) and one or more decay parameters are determined according to an exponential decay function.
15. One or more processors, comprising:
circuitry to use one or more neural networks to determine a location of one or more digital representations of one or more objects depicted within two or more first frames of a video based, at least in part, on motion of one or more labeled versions of the one or more digital representations of the one or more objects depicted within one or more second frames of the video, wherein the two or more first frames comprise a larger subset of frames of the video than the one or more second frames.
16. The one or more processors of claim 15 , wherein the circuitry is further to be configured to:
apply one or more decay parameters to prior state information maintained for the one or more neural networks and used for identifying the one or more digital representations, a weighting of the prior state information being reduced according to the one or more decay parameters.
17. The one or more processors of claim 16 , wherein the circuitry is further to be configured to:
store the prior state information external to the one or more neural networks and provide the prior state information to the one or more neural networks for each set of input to the one or more neural networks.
18. The one or more processors of claim 16 , wherein the circuitry is further to be configured to:
obtain a continuous stream of video data; and
provide individual frames of the video data to the one or more neural networks for use in identifying the one or more digital representations and updating the prior state information.
19. The one or more processors of claim 15 , wherein the circuitry is further to be configured to:
determine one or more decay parameters using a hyper-optimization process and a selected decay function.
20. The one or more processors of claim 15 , wherein the one or more neural networks are trained using sequences of sparse video data, individual sequences including at least one labeled version of the one or more objects depicted within one of the one or more second frames of the video, and wherein the one or more second frames of the video are used to simulate state information and the two or more first frames of the video are used to update the state information and identify one or more digital representations of the one or more objects represented in the individual sequence.
21. The one or more processors of claim 15 , wherein the one or more neural networks include at least one recurrent neural network (RNN) and one or more decay parameters are determined according to an exponential decay function.
22. A system, comprising:
one or more processors to calculate parameters corresponding to one or more neural networks to determine a location of one or more unlabeled versions of one or more objects depicted within two or more first frames of a video based, at least in part, on motion of one or more labeled versions of the one or more objects depicted within one or more second frames of the video, wherein the two or more first frames comprise a larger subset of frames of the video than the one or more second frames; and
one or more memories to store the parameters.
23. The system of claim 22 , wherein the one or more processors are further to be configured to:
apply one or more decay parameters to prior state information maintained for the one or more neural networks, a weighting of the prior state information being reduced according to the one or more decay parameters.
24. The system of claim 23 , wherein the one or more processors are further to be configured to:
store the prior state information external to the one or more neural networks and provide the prior state information to the one or more neural networks for each set of input to the one or more neural networks.
25. The system of claim 23 , wherein the one or more processors are further to be configured to:
obtain a continuous stream of video data; and
provide individual frames of the video data to the one or more neural networks for
use in identifying one or more digital representations of the one or more objects, as well as for use in updating the prior state information.
26. The system of claim 22 , wherein the one or more processors are further to be configured to:
determine one or more decay parameters using a hyper-optimization process and a selected decay function.
27. The system of claim 22 , wherein the one or more neural networks are trained using sequences of sparse video data, individual sequences including at least one labeled version of the one or more objects depicted within one of the one or more second frames of the video, and wherein the one or more second frames of the video are used to simulate state information and the two or more first frames of the video are used to update the state information and identify one or more digital representations of the one or more objects represented in the individual sequence.
28. The system of claim 22 , wherein the one or more neural networks include at least one recurrent neural network (RNN) and one or more decay parameters are determined according to an exponential decay function.
29. A vehicle, comprising:
a camera;
a drive assembly configured to maneuver the vehicle; and
a navigation system configured to provide navigation instructions to the drive
assembly, the navigation system providing video data captured by the camera to a one or more processors, comprising circuitry to use one or more neural networks to determine a location of one or more unlabeled versions of one or more digital representations of one or more objects depicted within two or more first frames of a video based, at least in part, on motion of one or more labeled versions of the one or more digital representations of the one or more objects depicted within one or more second frames of the video, wherein the two or more first frames comprise a larger subset of frames of the video than the one or more second frames.
30. The vehicle of claim 29 , wherein the one or more processors of the navigation system are further to be configured to:
apply one or more decay parameters to prior state information maintained for the one or more neural networks and used for identifying the one or more digital representations, a weighting of the prior state information being reduced according to the one or more decay parameters.
31. The vehicle of claim 30 , wherein the one or more processors of the navigation system are further to be configured to:
store the prior state information external to the one or more neural networks and provide the prior state information to the one or more neural networks for each set of input to the one or more neural networks.
32. The vehicle of claim 30 , wherein the one or more processors of the navigation system are further to be configured to:
obtain a continuous stream of video data; and
provide individual frames of the video data to the one or more neural networks for use in identifying the one or more digital representations and updating the prior state information.
33. The vehicle of claim 29 , wherein the one or more processors of the navigation system are further to be configured to:
determine one or more decay parameters using a hyper-optimization process and a selected decay function.
34. The vehicle of claim 29 , wherein the one or more neural networks are trained using sequences of sparse video data, individual sequences including at least one labeled version of the one or more objects depicted within one of the one or more second frames of the video, and wherein the one or more second frames of the video are used to simulate state information and the two or more first frames of the video are used to update the state information and identify one or more digital representations of the one or more objects represented in the individual sequence.
35. The vehicle of claim 29 , wherein the one or more neural networks include at least one recurrent neural network (RNN) and one or more decay parameters are determined according to an exponential decay function.