Pre-training model generation method, device and equipment for target detection tasks
By generating multiple input images and calculating feature representation comparison loss and positioning loss to train the model, the problem of poor results in migration of pre-trained models to downstream object detection tasks in the prior art is solved, and a more accurate object detection effect is achieved.
CN114202074BActive Publication Date: 2025-02-07BEIJING BAIDU NETCOM SCI & TECH CO LTD
Patent Information
- Application Number
- CN202111320361.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-11-09
- Publication Date
- 2025-02-07
- Estimated Expiration
- 2041-11-09
AI Technical Summary
Technical Problem
The pre-trained model generated by the existing comparative learning self-supervised algorithm is directly migrated to the downstream object detection task, resulting in poor object detection effect.
Method used
By acquiring the sample image and generating a plurality of input images, inputting to the preset first model and the second model, the feature representations of comparison loss and positioning loss are calculated, and the model is trained to generate a pretrained model. This pre-trained model can not only learn the ability to characterize the target, but also learn the ability to position the target.
Benefits of technology
Improve the accuracy of target detection results and improve the target detection effect.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN114202074B_ABST
Abstract
The present application discloses a method, device and equipment for generating a pre-trained model for target detection tasks, which relates to the field of artificial intelligence, and specifically to deep learning and computer vision technology. The specific implementation scheme is: obtaining a sample image, and generating a first input image, a second input image, a third input image and a fourth input image according to the sample image; generating target feature vectors of the first input image, the second input image, the third input image and the fourth input image according to multiple input images and the preset first model and second model; calculating feature representation contrast loss and positioning loss according to the target feature vectors of the first input image, the second input image, the third input image and the fourth input image; training the first model and the second model according to the feature representation contrast loss and positioning loss, and generating a pre-trained model according to the backbone network parameters of the trained first model.
Need to check novelty before this filing date? Find Prior Art
Citation Information
Patent Citations
Remote sensing image target detection method based on spatial hierarchy perception module and metric learning
CN111242026A
Target detection model training method and device, computer equipment and medium
CN112734641A