REGION OF INTEREST CONVOLUTIONAL NEURAL NETWORK PROCESSING

1 . A method for region of interest (ROI) convolutional neural network (CNN) processing, the method comprises:

applying, by the CNN, multiple CNN processing operations on input information received by the CNN, to provide one or more CNN output results;

wherein the applying comprises:

receiving, by a first intermediate CNN layer of the CNN, first intermediate information from a layer that precedes the first intermediate CNN layer;

applying, by the first intermediate CNN layer, a CNN processing operation only on first intermediate information included within a first ROI; and

preventing from applying the CNN processing operation on intermediate information outside the first ROI.

2 . The method according to claim 1 comprising receiving a definition of the first ROI.

3 . The method according to claim 1 wherein the applying comprises:

receiving, by a second intermediate CNN layer of the CNN, second intermediate information from a layer that precedes the second intermediate CNN layer; wherein the second intermediate CNN layer differs from the first intermediate CNN layer;

applying, by the second intermediate CNN layer, a CNN processing operation only on second intermediate information included within a second region of interest (ROI); and

preventing from applying the CNN processing operation on intermediate information outside the second ROI.

4 . The method according to claim 1 wherein the applying comprises generating the first intermediate information by one or more layers of the CNN that precede the first intermediate CNN layer.

5 . The method according to claim 1 wherein the applying comprises implementing CNN processing by different intermediate layers by reusing a convolutional module.

6 . The method according to claim 5 wherein the reusing comprises dynamically adjusting regions of interest between one use of the convolutional module to another.

7 . A non-transitory computer readable medium for region of interest (ROI) convolutional neural network (CNN) processing, the non-transitory computer readable medium stores instructions for:

applying, by the CNN, multiple CNN processing operations on input information received by the CNN, to provide one or more CNN output results;

wherein the applying comprises:

receiving, by a first intermediate CNN layer of the CNN, first intermediate information from a layer that precedes the first intermediate CNN layer;

applying, by the first intermediate CNN layer, a CNN processing operation only on first intermediate information included within a first ROI; and

preventing from applying the CNN processing operation on intermediate information outside the first ROI.

8 . The non-transitory computer readable medium according to claim 7 comprising receiving a definition of the first ROI.

9 . The non-transitory computer readable medium according to claim 7 wherein the applying comprises:

receiving, by a second intermediate CNN layer of the CNN, second intermediate information from a layer that precedes the second intermediate CNN layer; wherein the second intermediate CNN layer differs from the first intermediate CNN layer;

applying, by the second intermediate CNN layer, a CNN processing operation only on second intermediate information included within a second region of interest (ROI); and

preventing from applying the CNN processing operation on intermediate information outside the second ROI.

10 . The non-transitory computer readable medium according to claim 7 wherein the applying comprises generating the first intermediate information by one or more layers of the CNN that precede the first intermediate CNN layer.

11 . The non-transitory computer readable medium according to claim 7 wherein the applying comprises implementing CNN processing by different intermediate layers by reusing a convolutional module.

12 . The non-transitory computer readable medium according to claim 11 wherein the reusing comprises dynamically adjusting regions of interest between one use of the convolutional module to another.

13 . A neural network processor for region of interest (ROI) convolutional neural network (CNN) processing, the neural network processor either comprises a CNN or is configured to implement a CNN;

wherein the neural network processor is configured to apply multiple CNN processing operations on input information received by the neural network processor, to provide one or more CNN output results;

wherein the applying comprises:

receiving, by a first intermediate CNN layer of the CNN, first intermediate information from a layer that precedes the first intermediate CNN layer;

applying, by the first intermediate CNN layer, a CNN processing operation only on first intermediate information included within a first ROI; and

preventing from applying the CNN processing operation on intermediate information outside the first ROI.

14 . The neural network processor according to claim 13 that is configured to receive a definition of the first ROI.

15 . The neural network processor according to claim 14 that is configured to:

receive, by a second intermediate CNN layer of the CNN, second intermediate information from a layer that precedes the second intermediate CNN layer; wherein the second intermediate CNN layer differs from the first intermediate CNN layer;

apply, by the second intermediate CNN layer, a CNN processing operation only on second intermediate information included within a second region of interest (ROI); and

prevent from applying the CNN processing operation on intermediate information outside the second ROI.

16 . The neural network processor according to claim 13 that is configured to generate the first intermediate information by one or more layers of the CNN that precede the first intermediate CNN layer.

17 . The neural network processor according to claim 7 that is configured to implement CNN processing by different intermediate layers by reusing a convolutional module.

18 . The neural network processor according to claim 17 wherein the reusing comprises dynamically adjusting regions of interest between one use of the convolutional module to another.