Method and apparatus with object classification
An object classification method and apparatus are disclosed. The object classification method includes receiving an input image, storing first feature data extracted by a first feature extraction layer of a neural network configured to extract features of the input image, receiving second feature data from a second feature extraction layer which is an upper layer of the first feature extraction layer, generating merged feature data by merging the first feature data and the second feature data, and classifying an object in the input image based on the merged feature data.
1. A processor-implemented method, comprising:
receiving an input image;
generating first feature data corresponding to global information of the input image by extracting the first feature data using a first feature extraction layer of a neural network with the input image as input;
generating second feature data corresponding to local information of the input image by extracting the second feature data using a second feature extraction layer of the neural network, the second feature extraction layer being an upper layer of the first feature extraction layer;
generating sequentially merged feature data by merging the first feature data corresponding to the global information and sets of second feature data corresponding to local information of the input image output from a plurality of second feature extraction layers of the neural network; and
classifying an object in the input image based on the sequentially merged feature data.
2. The method of claim 1 , wherein the generating of the sequentially merged feature data comprising:
storing the sets of the second feature data corresponding to local features output from the plurality of second feature extraction layers.
3. The method of claim 1 , wherein the first feature data comprises one of a feature vector, a feature map, activation data and an activation map, and
the second feature data comprises one of a feature vector, a feature map, activation data and an activation map.
4. The method of claim 3 , wherein the generating of the sequentially merged feature data is based on an inner product between a feature map corresponding to the first feature data and a feature vector corresponding to the second feature data.
5. The method of claim 1 , wherein the second feature extraction layer is an uppermost feature extraction layer among a plurality of feature extraction layers comprised in the neural network.
6. The method of claim 1 , wherein the generating of the sequentially merged feature data comprises:
determining a weight to be applied to the first feature data;
applying the determined weight to the first feature data and determining first feature data to which the weight is applied; and
generating the sequentially merged feature data by merging the second feature data and the first feature data to which the weight is applied.
7. The method of claim 1 , wherein the generating of the sequentially merged feature data comprises:
converting the second feature data such that a dimension of the second feature data corresponds to a dimension of the first feature data; and
generating the sequentially merged feature data by merging the converted second feature data and the first feature data.
8. The method of claim 7 , wherein the generating of the sequentially merged feature data further comprises:
converting the sequentially merged feature data such that a dimension of the sequentially merged feature data corresponds to the dimension of the second feature data.
9. The method of claim 1 , wherein the first feature data comprises a local feature of the input image, and the second feature data comprises a global feature of the input image.
10. A non-transitory computer-readable storage medium storing instructions that, when executed by a processor, cause the processor to perform the method of claim 1 .
11. An electronic apparatus, comprising:
one or more processors, configured to:
receive an input image;
generate first feature data corresponding to global information of the input image by extracting the first feature data using a first feature extraction layer of a neural network with the input Image as input;
generate second feature data corresponding to local information of the input image by extracting the second feature data using a second feature extraction layer of the neural network, the second feature extraction layer being an upper layer of the first feature extraction layer;
generate sequentially merged feature data by merging the first feature data corresponding to the global information and sets of second feature data corresponding to local features-information of the input image output from a plurality of second feature extraction layers of the neural network; and
classify an object in the input image based on the sequentially merged feature data.
12. The electronic apparatus of claim 11 , wherein the processors are further configured to:
store the sets of the second feature data corresponding to local features output from the plurality of second feature extraction layers.
13. The electronic apparatus of claim 11 , wherein the first feature data comprises one of a feature vector, a feature map, activation data and an activation map, and
the second feature data comprises one of a feature vector, a feature map, activation data and an activation map.
14. The electronic apparatus of claim 13 , wherein the sequentially
merged feature data is based on an inner product between a feature map corresponding to the first feature data and a feature vector corresponding to the second feature data.
15. The electronic apparatus of claim 11 , wherein the second feature extraction layer is an uppermost feature extraction layer among a plurality of feature extraction layers comprised in the neural network.
16. The electronic apparatus of claim 11 , wherein the processors are further configured to:
determine a weight to be applied to the first feature data;
apply the determined weight to the first feature data and determining first feature data to which the weight is applied; and
generate the sequentially merged feature data by merging the second feature data and the first feature data to which the weight is applied.
17. The electronic apparatus of claim 11 , wherein the processors are further configured to:
convert the second feature data such that a dimension of the second feature data corresponds to a dimension of the first feature data; and
generate the sequentially merged feature data by merging the converted second feature data and the first feature data.
18. The electronic apparatus of claim 17 , wherein the processors are further configured to:
convert the sequentially merged feature data such that a dimension of the sequentially merged feature data corresponds to the dimension of the second feature data.