US 11,900,661 B2
Image processing method, device, storage medium and camera
Jiangtao Wen, Shenzhen (CN); Yuxing Han, Shenzhen (CN); Yanghao Li, Shenzhen (CN); Jiawen Gu, Shenzhen (CN); and Rui Zhang, Shenzhen (CN)
Assigned to BOYAN TECHNOLOGIES (SHENZHEN) CO., LTD, Shenzhen (CN)
Filed by BOYAN TECHNOLOGIES (SHENZHEN) CO., LTD, Shenzhen (CN)
Filed on Dec. 9, 2020, as Appl. No. 17/116,890.
Claims priority of application No. 202010085056.7 (CN), filed on Feb. 10, 2020.
Prior Publication US 2021/0250495 A1, Aug. 12, 2021
Int. Cl. G06V 10/82 (2022.01); G06N 3/04 (2023.01); G06F 18/213 (2023.01); H04N 23/80 (2023.01); G06V 10/764 (2022.01); G06V 10/77 (2022.01)
CPC G06V 10/82 (2022.01) [G06N 3/04 (2013.01); G06V 10/764 (2022.01); G06V 10/7715 (2022.01); H04N 23/80 (2023.01); G06F 18/213 (2023.01)] 7 Claims
OG exemplary drawing
 
1. An image processing method, applied to a camera, comprising:
capturing a target image;
acquiring a target feature map of the target image through a preset target convolution layer, wherein the target convolution layer includes at least one of a plurality of convolution layers of a convolutional neural network (CNN); and
outputting the target feature map;
wherein the target convolution layer is determined in the following ways:
acquiring a data volume of a feature map output by each of the plurality of convolution layers of the CNN, and determining the target convolution layer among the plurality of convolution layers based on the data volume; or
obtaining a configuration information of the camera, and determining the target convolution layer among the plurality of convolution layers of the CNN according to the configuration information;
wherein the configuration information comprises a first memory space corresponding to the camera, and the determining the target convolution layer among the plurality of convolution layers of the CNN according to the configuration information comprises:
obtaining a second memory space required by each of the plurality of convolution layers; and
obtaining a first sum value of the second memory space from a first convolution layer to a Nth convolution layer among the plurality of convolution layers, wherein if the first sum value is less than or equal to the first memory space and a second sum value of the second memory space from the first convolution layer to a N+1th convolution layer is greater than the first memory space, the Nth convolution layer and the previous convolution layer(s) of the Nth convolution layer among the plurality of convolution layers are used as the target convolution layers, where N is the integer greater than 0.