Cascade convolutional neural network
In one embodiment, an apparatus comprises a communication interface and a processor. The communication interface is to communicate with a plurality of devices. The processor is to: receive compressed data from a first device, wherein the compressed data is associated with visual data captured by sensor(s); perform a current stage of processing on the compressed data using a current CNN, wherein the current stage of processing corresponds to one of a plurality of processing stages associated with the visual data, and wherein the current CNN corresponds to one of a plurality of CNNs associated with the plurality of processing stages; obtain an output associated with the current stage of processing; determine, based on the output, whether processing associated with the visual data is complete; if the processing is complete, output a result associated with the visual data; if the processing is incomplete, transmit the compressed data to a second device.
1. A computing device to perform video content detection, comprising:
interface circuitry; and
processing circuitry to:
receive, via the interface circuitry, compressed video data, wherein the compressed video data represents video data in a compressed format; and
detect video content in the compressed video data using a cascade convolutional neural network (CNN), wherein:
the cascade CNN comprises a set of CNNs, wherein each CNN in the set of CNNs is trained to perform video content detection based on a corresponding set of video features encoded within the compressed video data; and
the cascade CNN performs video content detection using one or more CNNs from the set of CNNs successively until the video content in the compressed video data is detected.
2. The computing device of claim 1 , wherein the set of CNNs comprises:
one or more compressed-domain CNNs trained to perform video content detection based on compressed-domain features encoded within the compressed video data; and
one or more pixel-domain CNNs trained to perform video content detection based on pixel-domain features encoded within the compressed video data.
3. The computing device of claim 2 , wherein the compressed-domain features comprise:
motion vectors;
motion prediction residuals;
transform coefficients;
quantization parameters; or
macroblock coding modes.
4. The computing device of claim 2 , wherein:
the one or more compressed-domain CNNs comprise:
a first CNN trained to perform video content detection based on a first set of compressed-domain features encoded within the compressed video data; and
a second CNN trained to perform video content detection based on a second set of compressed-domain features encoded within the compressed video data; and
the one or more pixel-domain CNNs comprise:
a third CNN trained to perform video content detection based on pixel data encoded within the compressed video data.
5. The computing device of claim 4 , wherein the processing circuitry to detect the video content in the compressed video data using the cascade CNN is further to:
perform video content detection using the first CNN based on the first set of compressed-domain features;
upon determining that the first CNN is unable to detect the video content in the compressed video data, perform video content detection using the second CNN based on the second set of compressed-domain features; and
upon determining that the second CNN is unable to detect the video content in the compressed video data, perform video content detection using the third CNN based on the pixel data.
6. The computing device of claim 4 , wherein:
the first set of compressed-domain features comprises motion vectors encoded in the compressed video data; and
the second set of compressed-domain features comprises motion prediction residuals encoded in the compressed video data.
7. The computing device of claim 1 , wherein the computing device is an edge processing device for performing video analytics.
8. The computing device of claim 1 , wherein the compressed video data is compressed based on a motion-compensated video codec.
9. The computing device of claim 8 , wherein the motion-compensated video codec comprises an H.264 video codec.
10. 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, via interface circuitry, compressed video data, wherein the compressed video data represents video data in a compressed format; and
detect video content in the compressed video data using a cascade convolutional neural network (CNN), wherein:
the cascade CNN comprises a set of CNNs, wherein each CNN in the set of CNNs is trained to perform video content detection based on a corresponding set of video features encoded within the compressed video data; and
the cascade CNN performs video content detection using one or more CNNs from the set of CNNs successively until the video content in the compressed video data is detected.
11. The storage medium of claim 10 , wherein the set of CNNs comprises:
one or more compressed-domain CNNs trained to perform video content detection based on compressed-domain features encoded within the compressed video data; and
one or more pixel-domain CNNs trained to perform video content detection based on pixel-domain features encoded within the compressed video data.
12. The storage medium of claim 11 , wherein the compressed-domain features comprise:
motion vectors;
motion prediction residuals;
transform coefficients;
quantization parameters; or
macroblock coding modes.
13. The storage medium of claim 11 , wherein:
the one or more compressed-domain CNNs comprise:
a first CNN trained to perform video content detection based on a first set of compressed-domain features encoded within the compressed video data; and
a second CNN trained to perform video content detection based on a second set of compressed-domain features encoded within the compressed video data; and
the one or more pixel-domain CNNs comprise:
a third CNN trained to perform video content detection based on pixel data encoded within the compressed video data.
14. The storage medium of claim 13 , wherein the instructions that cause the processing circuitry to detect the video content in the compressed video data using the cascade CNN further cause the processing circuitry to:
perform video content detection using the first CNN based on the first set of compressed-domain features;
upon determining that the first CNN is unable to detect the video content in the compressed video data, perform video content detection using the second CNN based on the second set of compressed-domain features; and
upon determining that the second CNN is unable to detect the video content in the compressed video data, perform video content detection using the third CNN based on the pixel data.
15. The storage medium of claim 13 , wherein:
the first set of compressed-domain features comprises motion vectors encoded in the compressed video data; and
the second set of compressed-domain features comprises motion prediction residuals encoded in the compressed video data.
16. The storage medium of claim 10 , wherein the compressed video data is compressed based on a motion-compensated video codec.
17. The storage medium of claim 10 , wherein the motion-compensated video codec comprises an H.264 video codec.
18. A method of performing video content detection, comprising:
receiving, via interface circuitry, compressed video data, wherein the compressed video data represents video data in a compressed format; and
detecting video content in the compressed video data using a cascade convolutional neural network (CNN), wherein:
the cascade CNN comprises a set of CNNs, wherein each CNN in the set of CNNs is trained to perform video content detection based on a corresponding set of video features encoded within the compressed video data; and
the cascade CNN performs video content detection using one or more CNNs from the set of CNNs successively until the video content in the compressed video data is detected.
19. The method of claim 18 , wherein the set of CNNs comprises:
one or more compressed-domain CNNs trained to perform video content detection based on compressed-domain features encoded within the compressed video data; and
one or more pixel-domain CNNs trained to perform video content detection based on pixel-domain features encoded within the compressed video data.
20. The method of claim 19 , wherein the compressed-domain features comprise:
motion vectors;
motion prediction residuals;
transform coefficients;
quantization parameters; or
macroblock coding modes.
21. The method of claim 19 , wherein:
the one or more compressed-domain CNNs comprise:
a first CNN trained to perform video content detection based on a first set of compressed-domain features encoded within the compressed video data; and
a second CNN trained to perform video content detection based on a second set of compressed-domain features encoded within the compressed video data; and
the one or more pixel-domain CNNs comprise:
a third CNN trained to perform video content detection based on pixel data encoded within the compressed video data.
22. The method of claim 21 , wherein detecting the video content in the compressed video data using the cascade CNN comprises:
performing video content detection using the first CNN based on the first set of compressed-domain features;
upon determining that the first CNN is unable to detect the video content in the compressed video data, performing video content detection using the second CNN based on the second set of compressed-domain features; and
upon determining that the second CNN is unable to detect the video content in the compressed video data, performing video content detection using the third CNN based on the pixel data.
23. The method of claim 21 , wherein:
the first set of compressed-domain features comprises motion vectors encoded in the compressed video data; and
the second set of compressed-domain features comprises motion prediction residuals encoded in the compressed video data.
24. A system for performing video analytics, comprising:
a camera to capture video data;
video encoder circuitry to compress the video data into compressed video data, wherein the compressed video data represents the video data in a compressed format; and
processing circuitry to:
receive the compressed video data from the video encoder circuitry;
detect video content in the compressed video data using a cascade convolutional neural network (CNN), wherein:
the cascade CNN comprises a set of CNNs, wherein each CNN in the set of CNNs is trained to perform video content detection based on a corresponding set of video features encoded within the compressed video data; and
the cascade CNN performs video content detection using one or more CNNs from the set of CNNs successively until the video content in the compressed video data is detected.
25. The system of claim 24 , wherein the system is comprised in a smart camera.