Image compression with bounded deep neural network perception loss

1. A method comprising:

transmitting a plurality of probe images from an Internet of Things (IoT) device at an edge network to a server hosting a target deep neural network (DNN), wherein the plurality of images are injected with a limited amount of noise;

receiving a feedback comprising a plurality of discrete cosine transform (DCT) coefficients from the server hosting the target DNN, wherein the plurality of DCT coefficients are unique to the target DNN;

generating a quantization table based on the feedback received from the server hosting the target DNN;

compressing a set of real-time images using the generated quantization table by the IoT device at the edge network; and

transmitting the compressed set of real-time images to the server hosting the target DNN for DNN inferences.

2. The method of claim 1 , wherein the plurality of DCT coefficient indicates the DCT frequency domain sensitivity corresponding to the target DNN.

3. The method of claim 1 , further comprising:

converting the plurality of probe images from a spatial domain to a frequency domain prior to transmitting the plurality of probe images from the IoT device to the server hosting the target DNN.

4. The method of claim 1 , further comprising:

receiving an upper bound quota for loss increase associated with the target DNN, wherein the upper bound controls balance between file size and inference accuracy of the target DNN;

allocating the upper bound quota evenly among the plurality of DCT coefficients;

computing a corresponding upper bound of quantization noise on each of the plurality of DCT coefficients; and

generating the quantization table based on the corresponding upper bound of quantization noise on each DCT coefficient.

5. The method of claim 4 , wherein the corresponding upper bound of the quantization noise on each of the plurality of DCT coefficients is computed as allocated upper bound quota of loss increase divided by a gradient of loss with respect to each DCT coefficient.

6. The method of claim 5 , further comprising:

estimating the gradient of loss with respect to each DCT coefficient by performing backward propagation of loss.

7. The method of claim 1 , wherein the quantization table is generated during an offline phase when the IoT device is not concurrently performing real-time compression of the set of images at the edge network.

8. The method of claim 1 , wherein using the generated quantization table to perform the real-time compression of images captured by the IoT device at the edge network further comprises:

applying DCT on a live image by the IoT device to generate a plurality of DCT coefficient;

using the quantization table to generate quantized coefficients; and

performing entropy coding on the quantized coefficient to produce compressed image that are sensitive to the target DNN.

9. A network device comprising at least:

a memory;

a processor executing instructions in the memory to:

transmit a plurality of probe images from an Internet of Things (IoT) device at an edge network to a server hosting a target deep neural network (DNN), wherein the plurality of images are injected with a limited amount of noise;

receive a feedback comprising a plurality of discrete cosine transform (DCT) coefficients from the server hosting the target DNN, wherein the plurality of DCT coefficients are unique to the target DNN;

generate a quantization table based on the feedback received from the server hosting the target DNN;

compress a set of real-time images using the generated quantization table by the IoT device at the edge network; and

transmit the compressed set of real-time images to the server hosting the target DNN for DNN inferences.

10. The network device of claim 9 , wherein the plurality of DCT coefficient indicates the DCT frequency domain sensitivity corresponding to the target DNN.

11. The network device of claim 9 , wherein the processor further executes the instructions in the memory to:

convert the plurality of probe images from a spatial domain to a frequency domain prior to transmitting the plurality of probe images from the IoT device to the server hosting the target DNN.

12. The network device of claim 9 , wherein the processor further executes the instructions in the memory to:

receive an upper bound quota for loss increase associated with the target DNN, wherein the upper bound controls balance between file size and inference accuracy of the target DNN;

allocate the upper bound quota evenly among the plurality of DCT coefficients;

compute a corresponding upper bound of quantization noise on each of the plurality of DCT coefficients; and

generate the quantization table based on the corresponding upper bound of quantization noise on each DCT coefficient.

13. The network device of claim 12 , wherein the corresponding upper bound of the quantization noise on each of the plurality of DCT coefficients is computed as allocated upper bound quota of loss increase divided by a gradient of loss with respect to each DCT coefficient.

14. The network device of claim 13 , wherein the processor further executes the instructions in the memory to:

estimate the gradient of loss with respect to each DCT coefficient by performing backward propagation of loss.

15. The network device of claim 9 , wherein the quantization table is generated during an offline phase when the IoT device is not concurrently performing real-time compression of the set of images at the edge network.

16. The network device of claim 9 , wherein the processor further executes the instructions in the memory to:

apply DCT on a live image by the IoT device to generate a plurality of DCT coefficient;

use the quantization table to generate quantized coefficients; and

perform entropy coding on the quantized coefficient to produce compressed image that are sensitive to the target DNN.

17. A non-transitory machine-readable storage medium encoded with instructions executable by a processor of a computing device, the non-transitory machine-readable storage medium comprising instructions to:

transmit a plurality of probe images from an Internet of Things (IoT) device at an edge network to a server hosting a target deep neural network (DNN), wherein the plurality of images are injected with a limited amount of noise;

receive a feedback comprising a plurality of discrete cosine transform (DCT) coefficients from the server hosting the target DNN, wherein the plurality of DCT coefficients are unique to the target DNN;

generate a quantization table based on the feedback received from the server hosting the target DNN;

compress a set of real-time images using the generated quantization table by the IoT device at the edge network; and

transmit the compressed set of real-time images to the server hosting the target DNN for DNN inferences.

18. The non-transitory machine-readable storage medium of claim 17 , further comprising instructions to:

receive an upper bound quota for loss increase associated with the target DNN, wherein the upper bound controls balance between file size and inference accuracy of the target DNN;

allocate the upper bound quota evenly among the plurality of DCT coefficients;

compute a corresponding upper bound of quantization noise on each of the plurality of DCT coefficients; and

generate the quantization table based on the corresponding upper bound of quantization noise on each DCT coefficient.

19. The non-transitory machine-readable storage medium of claim 17 , further comprising instructions to:

apply DCT on a live image by the IoT device to generate a plurality of DCT coefficient;

use the quantization table to generate quantized coefficients; and

perform entropy coding on the quantized coefficient to produce compressed image that are sensitive to the target DNN.

20. The non-transitory machine-readable storage medium of claim 17 , wherein the quantization table is generated during an offline phase when the IoT device is not concurrently performing real-time compression of the set of images at the edge network.