Shape fusion for image analysis
Various types of image analysis benefit from a multi-stream architecture that allows the analysis to consider shape data. A shape stream can process image data in parallel with a primary stream, where data from layers of a network in the primary stream is provided as input to a network of the shape stream. The shape data can be fused with the primary analysis data to produce more accurate output, such as to produce accurate boundary information when the shape data is used with semantic segmentation data produced by the primary stream. A gate structure can be used to connect the intermediate layers of the primary and shape streams, using higher level activations to gate lower level activations in the shape stream. Such a gate structure can help focus the shape stream on the relevant information and reduces any additional weight of the shape stream.
1. A computer-implemented method, comprising:
inferring boundary data for one or more objects represented in an image,
inferring, at least partially in parallel with inferring the boundary data, semantic data for the image; and
fusing the boundary data with the semantic data to infer a segmentation and an object type corresponding to at least one object of the one or more objects.
2. The computer-implemented method of claim 1 , further comprising:
inferring the boundary data in parallel with a primary stream inferring the semantic data using a multi-stream convolutional neural network (CNN).
3. The computer-implemented method of claim 2 , further comprising:
fusing the boundary data with the semantic data at one or more layers of the CNN.
4. The computer-implemented method of claim 2 , wherein intermediate layers of the CNN are connected using a plurality of gated convolutional layers.
5. The computer-implemented method of claim 2 , further comprising:
inferring boundary data from successive convolutional layers of a shape stream by identifying and removing portions of the image that do not correspond to the boundary data.
6. The computer-implemented method of claim 1 , further comprising:
fusing the boundary data with the semantic data using layers of a fusion network.
7. The computer-implemented method of claim 1 , further comprising:
inferring the segmentation for performing at least one voice activated input command based on a confidence value associated with a neural network and the at least one object of the one or more objects.
8. The computer-implemented method of claim 1 , further comprising:
inferring the segmentation using deep learning accelerators (“DLA”) for an advanced driver assistance system (“ADAS”), an autonomous driving system, augmented reality (“AR) applications, and virtual reality (“VR”) applications.
9. The computer-implemented method of claim 1 , further comprising:
inferring the semantic data for the image based on a trained deep neural network using a subset of training data from a simulation.
10. The computer-implemented method of claim 1 , further comprising:
determining at least one action, based on the boundary data, the action associated with at least one of an autonomous vehicle, a robotic arm, and a security alarm.
11. A processor, comprising:
one or more circuits to:
receive image data captured of an environment;
infer boundary data for one or more objects represented in the image data,
infer, at least partially in parallel with inferring the boundary data, semantic data for the image data; and
fuse the boundary data with semantic data to infer a segmentation and an object type corresponding to at least one object of the one or more objects.
12. The processor of claim 11 , wherein the one or more circuits are further to:
infer the boundary data in parallel with a primary stream inferring the semantic data using a multi-stream convolutional neural network (CNN).
13. The processor of claim 11 , wherein the one or more circuits are further to:
infer the segmentation for performing at least one voice activated input command based on a confidence value associated with a neural network and the at least one object of the one or more objects.
14. The processor of claim 11 , wherein the one or more circuits are further to:
infer the segmentation using deep learning accelerators (“DLA”) for an advanced driver assistance system (“ADAS”), an autonomous driving system, augmented reality (“AR) applications, and virtual reality (“VR”) applications.
15. The processor of claim 11 , wherein the one or more circuits are further to:
infer the semantic data for the image based on a trained deep neural network using a subset of training data from a simulation.
16. The processor of claim 11 , wherein the one or more circuits are further to:
determine at least one action, based on the boundary data, the action associated with at least one of an autonomous vehicle, a robotic arm, and a security alarm.
17. A control system for an automated device, comprising:
a camera;
a control mechanism;
at least one processor; and
memory including instructions that, when executed by the at least one processor, cause the control system to:
capture by the camera, at least one image of an environment associated with the automated device;
infer boundary data of one or more objects represented in the at least one image,
infer, at least partially in parallel with inferring the boundary data, semantic data for the at least one image; and
fuse the boundary data with the semantic data to infer a segmentation and an object type corresponding to at least one object of the one or more objects.
18. The control system of claim 17 , wherein the instructions when executed further cause the system to:
infer the boundary data in parallel with a primary stream which infers the semantic data using a convolutional neural network (CNN).
19. The control system of claim 17 , wherein the instructions when executed further cause the system to:
infer the segmentation for performing at least one voice activated input command based on a confidence value associated with a neural network and the at least one object of the one or more objects.
20. The control system of claim 17 , wherein the instructions when executed further cause the system to:
adjust, based on the inferred segmentation, a path of navigation associated with a current location and a current direction of the automated device.