Entropy-based pre-filtering using neural networks for streaming applications

A system comprising:

• one or more processing units to execute operations comprising:
• determining a value for an entropy control parameter based at least on analyzing at least one of network or system conditions associated with an end-user device;
• selecting a deep neural network (DNN) from a plurality of DNNs based at least on the DNN including a respective value corresponding to the entropy control parameter;
• generating, using at least one of surface normals or depth values corresponding to a frame, a saliency map indicative of edges depicted in the frame;
• computing, using the selected DNN and based at least on data representative of the saliency map and the frame, a pre-filtered frame;
• and
• transmitting data representative of the pre-filtered frame to an end-user device.

A method comprising:

• selecting a value of an entropy control parameter based at least on one or more network or system conditions associated with an end-user device, wherein respective values for the entropy control parameter correspond to respective entropy levels;
• generating, using at least one of surface normal information corresponding to a frame or depth information corresponding to a frame, a saliency map indicative of edges depicted in the frame;
• based at least on the selecting of the value of the entropy control parameter, filtering, using a deep neural network (DNN) and the saliency map, the frame using an entropy level of the respective entropy levels that corresponds to the value to produce a pre-filtered frame;
• and
• transmitting data representative of the pre-filtered frame to an end-user device.