System and method for a deep learning machine for object detection
Apparatuses and methods of manufacturing same, systems, and methods for object detection using a region-based deep learning model are described. In one aspect, a method is provided, in which a region proposal network (RPN) is used to identify regions of interest (RoI) in an image by assigning a confidence levels, the assigned confidence levels of the RoIs are used to boost the background score assigned by the downstream classifier to each RoI, and the background scores are used in a softmax function to calculate the final class probabilities for each object class.
1. A method of object detection in an image using a region-based deep learning model, the method comprising:
identifying, using a region proposal network (RPN), regions of interest (RoI) in the image and assigning a confidence levels to each identified RoI;
boosting a background score assigned by a downstream classifier to each RoI, using the confidence level assigned to the ROI and optical flow magnitude;
using the boosted background scores in a softmax function to calculate final class probabilities; and
identifying each RoI as including an object in a foreground of the image or as a part of the background of the image, based on the final class probabilities.
2. The method of claim 1 , wherein the object includes a pedestrian.
3. The method of claim 1 , wherein the region-based deep learning model is a faster region-based convolutional neural network (R-CNN).
4. The method of claim 1 , wherein the region-based deep learning model is a region-based fully convolutional network (R-FCN).
5. The method of claim 1 , wherein the confidence levels comprise P B , which is a probability of the RoI being the background, and P F , which is a probability of the RoI including the object in the foreground.
6. The method of claim 5 , wherein s 0 is the background score assigned by the downstream classifier to a RoI boosted according to the formula:
s
0
=
{
s
0
if
P
B
<
P
F
P
B
·
s
0
P
F
otherwise
.
7. The method of claim 1 , wherein using the assigned confidence levels of the RoIs to boost the background score assigned by the downstream classifier to each RoI comprises:
iteratively refining the boosted background scores.
8. The method of claim 1 , wherein semantic segmentation masks are also used to boost the background score assigned by the downstream classifier to each RoI.
9. An apparatus capable of object detection using a region-based deep learning model, comprising:
one or more non-transitory computer-readable media; and
at least one processor which, when executing instructions stored on one or more non-transitory computer readable media, performs the steps of:
identifying, using a region proposal network (RPN), regions of interest (RoI) in the image and assigning a confidence levels to each identified RoI;
boosting a background score assigned by a downstream classifier to each RoI, using the confidence level assigned to the ROI and optical flow magnitude;
using the boosted background scores in a softmax function to calculate final class probabilities; and
identifying each RoI as including an object in a foreground of the image or as a part of the background of the image, based on the final class probabilities.
10. The apparatus of claim 9 , where the object includes a pedestrian.
11. The apparatus of claim 9 , wherein the region-based deep learning model is a faster region-based convolutional neural network (R-CNN).
12. The apparatus of claim 9 , wherein the region-based deep learning model is a region-based fully convolutional network (R-FCN).
13. The apparatus of claim 9 , wherein the confidence levels comprise P B , which is a probability of the RoI being the background, and P F , which is a probability of the RoI including the object in the foreground.
14. The apparatus of claim 13 , wherein so is the background score assigned by the downstream classifier to a RoI boosted according to the formula:
s
0
=
{
s
0
if
P
B
<
P
F
P
B
·
s
0
P
F
otherwise
.
15. The apparatus of claim 9 , wherein using the assigned confidence levels of the RoIs to boost the background score assigned by the downstream classifier to each RoI comprises:
iteratively refining the boosted background scores.
16. The apparatus of claim 9 , wherein semantic segmentation masks are also used to boost the background score assigned by the downstream classifier to each RoI.
17. A method, comprising:
manufacturing a chipset comprising:
at least one processor which, when executing instructions stored on one or more non-transitory computer readable media, performs the steps of:
identifying, using a region proposal network (RPN), regions of interest (RoI) in the image and assigning a confidence levels to each identified RoI;
boosting a background score assigned by a downstream classifier to each RoI, using the confidence level assigned to the ROI and optical flow magnitude;
using the boosted background scores in a softmax function to calculate final class probabilities; and
identifying each RoI as including an object in a foreground of the image or as a part of the background of the image, based on the final class probabilities, and
the one or more non-transitory computer-readable media which store the instructions.
18. A method of testing an apparatus, comprising:
testing whether the apparatus has at least one processor which, when executing instructions stored on one or more non-transitory computer readable media, performs the steps of:
identifying, using a region proposal network (RPN), regions of interest (RoI) in the image and assigning a confidence levels to each identified RoI;
boosting a background score assigned by a downstream classifier to each RoI, using the confidence level assigned to the ROI and optical flow magnitude;
using the boosted background scores in a softmax function to calculate final class probabilities; and
identifying each RoI as including an object in a foreground of the image or as a part of the background of the image, based on the final class probabilities, and
testing whether the apparatus has the one or more non-transitory computer-readable media which store the instructions.