US 12,387,469 B2
Object detection model training apparatus, method, and non-transitory computer readable storage medium thereof
Chun-Chang Wu, Hsinchu (TW); and Shih-Tse Chen, Hsinchu (TW)
Assigned to Realtek Semiconductor Corporation, Hsinchu (TW)
Filed by Realtek Semiconductor Corporation, Hsinchu (TW)
Filed on Jul. 7, 2022, as Appl. No. 17/811,303.
Claims priority of application No. 110136184 (TW), filed on Sep. 29, 2021.
Prior Publication US 2023/0096697 A1, Mar. 30, 2023
Int. Cl. G06V 10/774 (2022.01); G06N 20/00 (2019.01); G06V 20/64 (2022.01)
CPC G06V 10/7747 (2022.01) [G06N 20/00 (2019.01); G06V 20/64 (2022.01)] 17 Claims
OG exemplary drawing
 
1. An object detection model training apparatus, comprising:
a storage, being configured to store a student model and a plurality of teacher models, wherein the teacher models at least comprise a first teacher model and a second teacher model;
a transceiver interface; and
a processor, being electrically connected to the storage and the transceiver interface, and being configured to perform following operations:
receiving a plurality of training images from the transceiver interface, wherein the training images correspond to an object category;
performing a first object detection of the object category on the training images to generate a piece of first label information corresponding to each of the training images by the first teacher model, wherein each of the piece of first label information indicates at least one first object frame in each of the training images;
training the student model based on the training images and the first label information;
performing a second object detection of the object category on the training images to generate a piece of second label information corresponding to each of the training images by the second teacher model, wherein each of the piece of second label information indicates at least one second object frame in each of the training images; and
training the student model based on the training images and the second label information, wherein a second object labeled quantity of the second label information corresponding to each of the training images is not less than a first object labeled quantity of the first label information corresponding to each of the training images;
wherein the storage further stores a third teacher model, and the processor further performs following operations:
performing a third object detection of the object category on the training images to generate a piece of third label information corresponding to each of the training images by the third teacher model; and
training the student model based on the training images and the third label information, wherein a third object labeled quantity of the third label information corresponding to each of the training images is not less than the second object labeled quantity of the second label information corresponding to each of the training images.