Frequency-domain convolutional neural network
In one embodiment, an apparatus comprises a memory and a processor. The memory is to store visual data associated with a visual representation captured by one or more sensors. The processor is to: obtain the visual data associated with the visual representation captured by the one or more sensors, wherein the visual data comprises uncompressed visual data or compressed visual data; process the visual data using a convolutional neural network (CNN), wherein the CNN comprises a plurality of layers, wherein the plurality of layers comprises a plurality of filters, and wherein the plurality of filters comprises one or more pixel-domain filters to perform processing associated with uncompressed data and one or more compressed-domain filters to perform processing associated with compressed data; and classify the visual data based on an output of the CNN.
1. At least one non-transitory computer-readable storage medium having instructions stored thereon, wherein the instructions, when executed on processing circuitry, cause the processing circuitry to:
receive a frequency-domain representation of image data, wherein the frequency-domain representation comprises a plurality of sets of transform coefficients for a plurality of blocks of the image data; and
detect content in the image data using a convolutional neural network (CNN), wherein the CNN is trained to detect the content in the image data based on corresponding transform coefficients among the plurality of blocks.
2. The storage medium of claim 1 , wherein the corresponding transform coefficients comprise groups of transform coefficients for corresponding frequencies from the plurality of sets of transform coefficients.
3. The storage medium of claim 1 , wherein:
the frequency-domain representation of the image data is arranged into a plurality of dimensions, wherein the plurality of dimensions comprise a first dimension, a second dimension, and a third dimension;
the first dimension and the second dimension represent spatial indices of the plurality of blocks within the image data; and
the third dimension represents a plurality of channels, wherein each channel comprises a group of transform coefficients for a corresponding frequency from the plurality of sets of transform coefficients.
4. The storage medium of claim 1 , wherein each set of transform coefficients represents a transform performed on one of the plurality of blocks.
5. The storage medium of claim 4 , wherein:
the transform is a discrete cosine transform (DCT); and
the transform coefficients are DCT coefficients.
6. The storage medium of claim 1 , wherein the CNN comprises one or more butterfly layers, wherein the one or more butterfly layers are to perform one or more partial inverse transforms based on the frequency-domain representation of the image data.
7. The storage medium of claim 1 , wherein the instructions that cause the processing circuitry to receive the frequency-domain representation of the image data further cause the processing circuitry to:
receive, via interface circuitry, compressed image data, wherein the compressed image data comprises the frequency-domain representation of the image data; and
extract the frequency-domain representation of the image data from the compressed image data.
8. The storage medium of claim 1 , wherein the image data comprises an image or a video frame.
9. A device, comprising:
processing circuitry to:
receive a frequency-domain representation of image data, wherein the frequency-domain representation comprises a plurality of sets of transform coefficients for a plurality of blocks of the image data; and
detect content in the image data using a convolutional neural network (CNN), wherein the CNN is trained to detect the content in the image data based on corresponding transform coefficients among the plurality of blocks.
10. The device of claim 9 , wherein the corresponding transform coefficients comprise groups of transform coefficients for corresponding frequencies from the plurality of sets of transform coefficients.
11. The device of claim 9 , wherein:
the frequency-domain representation of the image data is arranged into a plurality of dimensions, wherein the plurality of dimensions comprise a first dimension, a second dimension, and a third dimension;
the first dimension and the second dimension represent spatial indices of the plurality of blocks within the image data; and
the third dimension represents a plurality of channels, wherein each channel comprises a group of transform coefficients for a corresponding frequency from the plurality of sets of transform coefficients.
12. The device of claim 9 , wherein each set of transform coefficients represents a transform performed on one of the plurality of blocks.
13. The device of claim 12 , wherein:
the transform is a discrete cosine transform (DCT); and
the transform coefficients are DCT coefficients.
14. The device of claim 9 , wherein the CNN comprises one or more butterfly layers, wherein the one or more butterfly layers are to perform one or more partial inverse transforms based on the frequency-domain representation of the image data.
15. The device of claim 9 , wherein the processing circuitry to receive the frequency-domain representation of the image data is further to:
receive, via interface circuitry, compressed image data, wherein the compressed image data comprises the frequency-domain representation of the image data; and
extract the frequency-domain representation of the image data from the compressed image data.
16. The device of claim 9 , wherein the device is:
a smart camera; or
an edge server.
17. A method, comprising:
receiving a frequency-domain representation of image data, wherein the frequency-domain representation comprises a plurality of sets of transform coefficients for a plurality of blocks of the image data, wherein each set of transform coefficients represents a transform performed on one of the plurality of blocks; and
detecting content in the image data using a convolutional neural network (CNN), wherein the CNN is trained to detect the content in the image data based on corresponding transform coefficients among the plurality of blocks.
18. The method of claim 17 , wherein the corresponding transform coefficients comprise groups of transform coefficients for corresponding frequencies from the plurality of sets of transform coefficients.
19. The method of claim 17 , wherein:
the frequency-domain representation of the image data is arranged into a plurality of dimensions, wherein the plurality of dimensions comprise a first dimension, a second dimension, and a third dimension;
the first dimension and the second dimension represent spatial indices of the plurality of blocks within the image data; and
the third dimension represents a plurality of channels, wherein each channel comprises a group of transform coefficients for a corresponding frequency from the plurality of sets of transform coefficients.
20. The method of claim 17 , wherein:
the transform is a discrete cosine transform (DCT); and
the transform coefficients are DCT coefficients.