Cooperative saliency detection method based on image group adaptation
A detection method and image group technology, applied in the direction of neural learning methods, instruments, biological neural network models, etc., can solve the problems of lack of adaptive adjustment ability of different image groups, increase of image preprocessing time, and inconformity with actual scene requirements, etc. , to achieve the effect of low cost, good detection effect and easy realization
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0062] In this example, see figure 1 , a collaborative saliency detection method based on image group adaptation, the steps are as follows:
[0063] (1) Construct an image group adaptive collaborative saliency detection model:
[0064] (1-1) The model mainly includes a trained deep learning-based salient object detection model, a saliency correlation backbone network and a guidance sub-network, and the three networks jointly perform end-to-end collaborative saliency detection on image groups ;
[0065] (1-2) Prior knowledge acquisition of salient objects:
[0066] Determine a salient object detection model, which is any salient object detection model based on deep learning with trained parameters. The salient object detection model is used to generate salient object detection results of any image group and salient shallow feature extraction, as The saliency-related backbone network provides prior knowledge;
[0067] (1-3) Build a bootstrap subnet:
[0068] The network use...
Embodiment 2
[0080] This embodiment is basically the same as the first embodiment, and the special features are:
[0081] In this example, see figure 1 , in the step (1-3), the salient object result output by the salient object detection network and the salient shallow feature extracted by the salient object detection network are multiplied pixel by pixel, and after eliminating the interference of the non-salient area in the shallow feature, the feature is used as The input of the saliency correlation backbone network.
[0082] In this embodiment, the attention correlation module in the step (1-4), the module directly splices all the image features of a batch of input into an overall feature by length, and then performs spatial self-attention on the feature. Force and channel self-attention, add the features processed by the two attention mechanisms and restore them to the feature size of the original batch. This splicing method makes the number of images input to the model random, avoidi...
Embodiment 3
[0089] This embodiment is basically the same as the above-mentioned embodiment, and the special features are:
[0090] In this example, see figure 1 and figure 2 , a collaborative saliency detection method based on image group adaptation, the steps are as follows:
[0091] (1) Saliency prior knowledge acquisition: use any deep learning-based salient object detection model with trained parameters to generate any image group salient object detection results for and extract salient shallow features Provide prior knowledge for the saliency-related backbone network;
[0092] (2) Constructing the guidance sub-network GNet: The network uses VGG16 as the backbone network, which consists of 5 convolution blocks of VGG, an average pooling layer, three fully connected layers and two Relu layers; the input and salient object detection model The input is the same, and then the average pooling operation is applied to the features after 5 convolution blocks to obtain the feature vec...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More