Deep geometric model fitting
Systems, apparatuses and methods may provide for technology that generates, by a first neural network, an initial set of model weights based on input data and iteratively generates, by a second neural network, an updated set of model weights based on residual data associated with the initial set of model weights and the input data. Additionally, the technology may output a geometric model of the input data based on the updated set of model weights. In one example, the first neural network and the second neural network reduce the dependence of the geometric model on the number of data points in the input data.
1. A semiconductor apparatus comprising:
one or more substrates; and
logic coupled to the one or more substrates, wherein the logic is implemented at least partly in one or more of configurable logic or fixed-functionality hardware logic, the logic coupled to the one or more substrates to:
generate, by a first neural network, an initial set of model weights based on input data,
iteratively generate, by a second neural network, an updated set of model weights based on residual data associated with the initial set of model weights and the input data, wherein the second neural network is to learn one or more regularities associated with the input data, and wherein the one or more regularities include motion patterns, and
output a geometric model of the input data based on the updated set of model weights, wherein the first neural network and the second neural network are to reduce a dependence of the geometric model on a number of data points in the input data.
2. The apparatus of claim 1 , wherein the updated set of model weights are to be generated further based on one or more of line fitting constraints or data correspondence likelihoods.
3. The apparatus of claim 1 , wherein the geometric model is to be one or more of a hyperplane fit model, a fundamental matrix estimation model or a homography estimation model.
4. The apparatus of claim 1 , wherein the logic coupled to the one or more substrates includes transistor channel regions that are positioned within the one or more substrates.
5. A system comprising:
a plurality of cameras to generate input data;
a memory coupled to the plurality of cameras, the memory to store the input data;
a semiconductor apparatus coupled to the memory, the semiconductor apparatus including one or more substrates and logic coupled to the one or more substrates, the logic coupled to the one or more substrates to:
generate, by a first neural network, an initial set of model weights based on the input data,
iteratively generate, by a second neural network, an updated set of model weights based on residual data associated with the initial set of model weights and the input data, and
output a geometric model of the input data based on the updated set of model weights; and
a processor to localize the plurality of cameras in a three-dimensional environment based on the geometric model.
6. The system of claim 5 , wherein the first neural network and the second neural network are to reduce a dependence of the geometric model on a number of data points in the input data.
7. The system of claim 5 , wherein the second neural network is to learn one or more regularities associated with the input data.
8. The system of claim 7 , wherein the one or more regularities are to include motion patterns.
9. The system of claim 5 , wherein the updated set of model weights are to be generated further based on one or more of line fitting constraints or data correspondence likelihoods.
10. The system of claim 5 , wherein the geometric model is to be one or more of a hyperplane fit model, a fundamental matrix estimation model or a homography estimation model.
11. A semiconductor apparatus comprising:
one or more substrates; and
logic coupled to the one or more substrates, wherein the logic is implemented at least partly in one or more of configurable logic or fixed-functionality hardware logic, the logic coupled to the one or more substrates to:
generate, by a first neural network, an initial set of model weights based on input data,
iteratively generate, by a second neural network, an updated set of model weights based on residual data associated with the initial set of model weights and the input data, and
output a geometric model of the input data based on the updated set of model weights, wherein one or more of the first neural network or the second neural network is to reduce a dependence of the geometric model on a number of data points in the input data.
12. The apparatus of claim 11 , wherein the first neural network and the second neural network are to reduce the dependence of the geometric model on the number of data points in the input data.
13. The apparatus of claim 11 , wherein the second neural network is to learn one or more regularities associated with the input data.
14. The apparatus of claim 13 , wherein the one or more regularities are to include motion patterns.
15. The apparatus of claim 11 , wherein the updated set of model weights are to be generated further based on one or more of line fitting constraints or data correspondence likelihoods.
16. The apparatus of claim 11 , wherein the geometric model is to be one or more of a hyperplane fit model, a fundamental matrix estimation model or a homography estimation model.
17. The apparatus of claim 11 , wherein the logic coupled to the one or more substrates includes transistor channel regions that are positioned within the one or more substrates.
18. A method comprising:
generating, by a first neural network, an initial set of model weights based on input data;
iteratively generating, by a second neural network, an updated set of model weights based on residual data associated with the initial set of model weights and the input data, wherein the second neural network is to identify one or more regularities associated with the input data, and wherein the one or more regularities include motion patterns; and
outputting a geometric model of the input data based on the updated set of model weights.
19. The method of claim 18 , wherein the first neural network and the second neural network reduce a dependence of the geometric model on a number of data points in the input data.
20. The method of claim 18 , wherein the second neural network learns the one or more regularities associated with the input data.
21. The method of claim 18 , wherein the updated set of model weights are generated further based on one or more of line fitting constraints or data correspondence likelihoods.
22. The method of claim 18 , wherein the geometric model is one or more of a hyperplane fit model, a fundamental matrix estimation model or a homography estimation model.
23. At least one non-transitory computer readable storage medium comprising a set of instructions, which when executed by a computing system, cause the computing system to perform the method of claim 18 .