Cascaded detection of facial attributes
This disclosure provides methods, devices, and systems for object detection in images. The present implementations more specifically relate to cascaded object detection techniques. In some aspects, an object detection system may obtain a first image and perform a first object detection operation on the first image. The object detection system may select a region of interest (ROI) in the first image based at least in part on the object detection operation, the ROI being selected before the object detection operation is completed. The object detection system may obtain a second image coinciding with the ROI of the first image, the second image having a higher resolution than the first image. The object detection system may selectively perform an inferencing operation on the second image based on a result of the object detection operation.
1 . A method for object detection in images, comprising:
obtaining a first image captured by an image capture component;
performing an object detection operation on the first image;
selecting a region of interest (ROI) in the first image based at least in part on a partial result of the object detection operation, the partial result being generated before the object detection operation is completed;
obtaining a second image captured by the image capture component and coinciding with the ROI of the first image, the second image having a higher resolution than the ROI of the first image;
generating a final result based on the completion of the object detection operation;
determining whether the final result of the object detection operation validates the partial result; and
selectively performing an inferencing operation on the second image based on whether the final result of the object detection operation validates the partial result.
2 . The method of claim 1 , wherein the selecting of the ROI in the first image comprises selecting the ROI in the first image based on the partial result of the object detection operation on the first image.
3 . The method of claim 1 , wherein the object detection operation is performed based on an object detection model configured to detect humans.
4 . The method of claim 1 , wherein the inferencing operation comprises at least one of a gaze detection operation, a person identification operation, or a motion detection operation.
5 . The method of claim 1 , wherein the second image is obtained prior to completion of the object detection operation.
6 . The method of claim 1 , wherein the selecting of the ROI in the first image comprises:
detecting an object of interest in the first image based on the object detection operation;
identifying a location of the object of interest in the first image; and
selecting a portion of the first image coinciding with the location of the object of interest as the ROI.
7 . The method of claim 1 , wherein the selecting of the ROI in the first image comprises:
identifying one or more regions of non-interest (RON) in the first image based on the object detection operation, the ROI including any portion of the first image that does not overlap the one or more regions of non-interest (RON).
8 . The method of claim 1 , wherein the selecting of the ROI in the first image comprises:
performing a motion tracking operation, with respect to an object of interest, on the first image; and
identifying the ROI based on the motion tracking operation.
9 . The method of claim 1 , further comprising:
obtaining a third image associated with a portion of the second image; and
selectively performing a second inferencing operation on the third image.
10 . The method of claim 9 , wherein the obtaining of the third image is responsive to the final result of the object detection operation.
11 . The method of claim 9 , wherein the third image has a higher resolution than the second image.
12 . The method of claim 1 , wherein the final result of the object detection operation validating the partial result includes the final result indicating that an object of interest is detected in the ROI.
13 . The method of claim 1 , further comprising discarding the second image responsive to the final result of the object detection operation indicating that no object of interest is detected in the ROI.
14 . An object detection system, comprising:
one or more processors; and
a memory coupled to the one or more processors, the memory storing instructions that, when executed by the one or more processors, cause the object detection system to:
obtain a first image captured by an image capture component;
perform a first object detection operation on the first image;
select a region of interest (ROI) in the first image based at least in part on a partial result of the object detection operation, the partial result being generated before the object detection operation is completed;
obtain a second image captured by the image capture component and coinciding with the ROI of the first image, the second image having a higher resolution than the first image;
generate a final result based on the completion of the object detection operation;
determine whether the final result of the object detection operation validates the partial result; and
selectively perform an inferencing operation on the second image based on whether the final result of the object detection operation validates the partial result.
15 . The object detection system of claim 14 , wherein execution of the instructions further causes the object detection system to select the ROI in the first image based on the partial result of the object detection operation on the first image.
16 . The object detection system of claim 14 , wherein the second image is obtained prior to completion of the object detection operation.
17 . The object detection system of claim 14 , wherein execution of the instructions further causes the object detection system to:
detect an object of interest in the first image based on the object detection operation;
identify a location of the object of interest in the first image; and
select a portion of the first image coinciding with the location of the object of interest as the ROI.
18 . The object detection system of claim 14 , wherein execution of the instructions further causes the object detection system to identify one or more regions of non-interest (RON) in the first image based on the object detection operation, the ROI including any portion of the first image that does not overlap the one or more regions of non-interest (RON).
19 . The object detection system of claim 14 , wherein execution of the instructions further causes the object detection system to:
perform a motion tracking operation, with respect to an object of interest, on the first image; and
identify the ROI based on the motion tracking operation.
20 . The object detection system of claim 14 , wherein execution of the instructions further causes the object detection system to discard the second image responsive to the final result of the object detection operation indicating that no object of interest is detected in the ROI.