Network, system and method for image processing
A network for image processing is provided, and more particularly, for coarse-to-fine recognition of image processing. The network includes a shared convolution layer, and a first subnet and a second subnet both subsequent to the shared convolution layer; the first subnet comprises a first skipping module comprising one or more skip-dense blocks iteratively stacked with one or more transition layers, a first pooling layer subsequent to the first skipping module, and a first classification layer subsequent to the first pooling layer; the second subnet comprises a second skipping module comprising one or more skip-dense blocks iteratively stacked with one or more layers, a second pooling layer subsequent to the second skipping module, and a second classification layer subsequent to the second pooling layer; and wherein a skip-dense block of the second subnet is selected to guide a transition layer of the first subnet, and the level of the guiding skip-dense block is deeper than the level of the guided transition layer. This network is also related to a system and a method thereof.
1. A network of image processing, especially for coarse-to-fine recognition, comprising:
a shared convolution layer; and
a first subnet and a second subnet both subsequent to the shared convolution layer;
wherein the first subnet comprises
a first skipping module comprising one or more skip-dense blocks iteratively stacked with one or more transition layers;
a first pooling layer subsequent to the first skipping module, and
a first classification layer subsequent to the first pooling layer;
wherein the second subnet comprises
a second skipping module comprising one or more skip-dense blocks iteratively stacked with one or more layers;
a second pooling layer subsequent to the second skipping module, and
a second classification layer subsequent to the second pooling layer; and
wherein a skip-dense block of the second subnet is selected to guide a transition layer of the first subnet, and the level of the guiding skip-dense block is deeper than the level of the guided transition layer.
2. The network of claim 1 , wherein the deepest skip-dense block of the second subnet is selected to guide the deepest transition layer of the first skipping subnet.
3. The network of claim 1 , wherein the number of the transition layers is one less than the number of skip-dense blocks in the first skipping module, and/or the number of the transition layers is one less than the number of skip-dense blocks in the second skipping module.
4. The network of claim 1 , wherein a first loss function module is added to the first subnets at its outputs, and a second loss function module is added to the second subnets at its outputs, while the parameters of the first and second loss function are different.
5. The network of claim 1 , wherein each said skip-dense block comprises one or more sub-blocks each having a gating module, a dense layer subsequent to the gating module, and a merge module subsequent to the dense layer while also short connected to the input of the sub-block.
6. The network of claim 1 , wherein the first and second classification layers are linear classifiers.
7. A system of image processing, especially for coarse-to-fine recognition, comprising an input device, a processor for processing the input data, and an output device for outputting the processed data; wherein the processor is configured to build a computing model comprising:
a shared convolution layer,
a first subnet and a second subnet both subsequent to the shared convolution layer;
wherein the first subnet comprises
a first skipping module comprising one or more skip-dense blocks iteratively stacked with one or more transition layers;
a first pooling layer subsequent to the first skipping module, and
a first classification layer subsequent to the first pooling layer,
wherein the second subnet comprises
a second skipping module comprising one or more skip-dense blocks iteratively stacked with one or more layers;
a second pooling layer subsequent to the second skipping module, and
a second classification layer subsequent to the second pooling layer,
wherein a skip-dense block of the second subnet is selected to guide a transition layer of the first subnet, and the level of the guiding skip-dense block is deeper than the level of the guided transition layer.
8. The system of claim 7 , wherein the deepest skip-dense block of the second subnet is selected to guide the deepest transition layer of the first skipping subnet.
9. The system of claim 7 , wherein the number of the transition layers is one less than the number of skip-dense blocks in the first skipping module, and/or the number of the transition layers is one less than the number of skip-dense blocks in the second skipping module.
10. The system of claim 7 , wherein a first loss function module is added to the first subnets at its outputs, and a second loss function module is added to the second subnets at its outputs, while the parameters of the first and second loss function are different.
11. The system of claim 7 , wherein each said skip-dense block comprises one or more sub-blocks each having a gating module, a dense layer subsequent to the gating module, and a merge module subsequent to the dense layer while also short connected to the input of the sub-block.
12. The system of claim 7 , wherein the first and second classification layers are linear classifiers.
13. A method of image processing, especially for coarse-to-fine recognition, comprising the steps of:
inputting input data to a shared convolution layer;
outputting an output of the shared convolution layer to a first subnet and a second subnet both subsequent to the shared convolution layer;
wherein the first subnet comprises
a first skipping module comprising one or more skip-dense blocks iteratively stacked with one or more transition layers;
a first pooling layer subsequent to the first skipping module, and
a first classification layer subsequent to the first pooling layer,
wherein the second subnet comprises
a second skipping module comprising one or more skip-dense blocks iteratively stacked with one or more layers;
a second pooling layer subsequent to the second skipping module, and
a second classification layer subsequent to the second pooling layer; and
selecting a skip-dense block of the second subnet to guide a transition layer of the first subnet, and the level of the guiding skip-dense block is deeper than the level of the guided transition layer.
14. The method of claim 13 , wherein the step of selecting includes selecting the deepest skip-dense block of the second subnet to guide the deepest transition layer of the first skipping subnet.
15. The method of claim 13 , wherein the number of the transition layers is one less than the number of skip-dense blocks in the first skipping module, and/or the number of the transition layers is one less than the number of skip-dense blocks in the second skipping module.
16. The method of claim 13 , wherein a first loss function module is added to the first subnets at its outputs, and a second loss function module is added to the second subnets at its outputs, while the parameters of the first and second loss function are different.
17. The method of claim 13 , wherein each skip-dense block comprises one or more sub-blocks each having a gating module, a dense layer subsequent to the gating module, and a merge module subsequent to the dense layer while also short connected to the input of the sub-block.
18. The method of claim 13 , wherein the first and second classification layers are linear classifiers.