System and method for semantic segmentation using hybrid dilated convolution (HDC)
A system and method for semantic segmentation using hybrid dilated convolution (HDC) are disclosed. A particular embodiment includes: receiving an input image; producing a feature map from the input image; performing a convolution operation on the feature map and producing multiple convolution layers; grouping the multiple convolution layers into a plurality of groups; applying different dilation rates for different convolution layers in a single group of the plurality of groups; and applying a same dilation rate setting across all groups of the plurality of groups.
1. A method for vehicular control, comprising:
producing a feature map from an input image;
producing multiple convolution layers by performing a convolution operation on the feature map;
grouping the multiple convolution layers into a plurality of groups;
applying different dilation rates for different convolution layers in a single group of the plurality of groups, wherein the different dilation rates do not have a common factor relationship; and
applying a same dilation rate setting across each of the plurality of groups,
wherein the method is used by a control subsystem to control a vehicle.
2. The method of claim 1 , wherein the different dilation rates comprise increasing dilation rates for the different convolution layers in the single group.
3. The method of claim 2 , wherein each group of the plurality of groups comprises convolution layers and the method further comprises applying dilation rates for the convolution layers within each group, wherein the dilation rates for the convolution layers within each group are identical to the different dilation rates.
4. The method of claim 1 , wherein the vehicle is an autonomous vehicle.
5. The method of claim 4 , wherein the autonomous vehicle has no driver.
6. The method of claim 1 , wherein the method is used in conjunction with a dense upsampling convolution (DUC) method.
7. The method of claim 6 , wherein the DUC method is integrated into a fully convolutional network framework.
8. The method of claim 6 , wherein the DUC method is performed at an original resolution, thereby enabling pixel-level decoding.
9. A device for vehicular control, comprising:
a processor; and
a memory that comprises instructions stored thereupon, wherein the instructions when executed by the processor configure the processor to:
produce a feature map from an input image;
produce multiple convolution layers by performing a convolution operation on the feature map;
group the multiple convolution layers into a plurality of groups;
apply different dilation rates for different convolution layers in a single group of the plurality of groups, wherein the different dilation rates do not have a common factor relationship; and
apply a same dilation rate setting across each of the plurality of groups,
wherein the instructions are used by a control subsystem to control a vehicle.
10. The method of claim 9 , wherein the different dilation rates comprise increasing dilation rates for the different convolution layers in the single group.
11. The method of claim 10 , wherein each group of the plurality of groups comprises convolution layers and further configure the processor to apply dilation rates for the convolution layers within each group, wherein the dilation rates for the convolution layers within each group are identical to the different dilation rates.
12. The method of claim 9 , wherein the vehicle is an autonomous vehicle.
13. The method of claim 12 , wherein the autonomous vehicle has no driver.
14. The method of claim 9 , wherein the instructions are used in conjunction with other instructions that configure the process to implement a dense upsampling convolution method.
15. A non-transitory computer readable program storage medium having code stored thereon, the code, when executed by a processor, causing the processor to implement a method for vehicular control, the method comprising:
producing a feature map from an input image;
producing multiple convolution layers by performing a convolution operation on the feature map;
grouping the multiple convolution layers into a plurality of groups;
applying different dilation rates for different convolution layers in a single group of the plurality of groups, wherein the different dilation rates do not have a common factor relationship; and
applying a same dilation rate setting across each of the plurality of groups,
wherein the method is used by a control subsystem to control a vehicle.
16. The non-transitory computer readable program storage medium of claim 15 , wherein the different dilation rates comprise increasing dilation rates for the different convolution layers in the single group.
17. The non-transitory computer readable program storage medium of claim 16 , wherein each group of the plurality of groups comprises convolution layers and the method further comprises applying dilation rates for the convolution layers within each group, wherein the dilation rates for the convolution layers within each group are identical to the different dilation rates.
18. The non-transitory computer readable program storage medium of claim 15 , wherein the vehicle is an autonomous vehicle without a driver.
19. The non-transitory computer readable program storage medium of claim 15 , wherein the method is used in conjunction with a dense upsampling convolution (DUC) method.
20. The non-transitory computer readable program storage medium of claim 19 , wherein the DUC method is performed at an original resolution, thereby enabling pixel-level decoding.