Video compression technique using a machine learning system

A computer-implemented method for decompressing a set of video frames partitioned into two subsets of different types of frames, a first type and a second type, the method comprising:

• decoding a first representation to reconstruct said first type of frames using a first stage decoder, wherein said first representation is generated by compressing said first type of frames of videos by a first stage encoder;
• training a generative model corresponding to a second stage decoder using said first type of frames, said reconstructed first type of frames, and a third representation by using a discriminator employed by a machine learning system, wherein said third representation is generated by compressing said reconstructed first type of frames by a second stage encoder;
• and
• generating reconstructed first and second types of frames by said generative model using soft edge information after training of said generative model, wherein said soft edge information is contained in a second representation generated by compressing said second type of frames of video by said second stage encoder.

A computer program product for compressing video, the computer program product comprising one or more computer readable storage mediums having program code embodied therewith, the program code comprising the programming instructions for:

• partitioning a set of video frames into two subsets of different types of frames, a first type and a second type;
• compressing said first type of frames of videos to generate a first representation by a first stage encoder;
• and
• compressing said second type of frames of video to generate a second representation that only contains soft edge information by a second stage encoder, wherein said soft edge information comprises a soft map at a particular quantization level, wherein said soft map comprises a map of multilevel edges.

A computer program product for decompressing a set of video frames partitioned into two subsets of different types of frames, a first type and a second type, the computer program product comprising one or more computer readable storage mediums having program code embodied therewith, the program code comprising the programming instructions for:

• decoding a first representation to reconstruct said first type of frames using a first stage decoder, wherein said first representation is generated by compressing said first type of frames of videos by a first stage encoder;
• training a generative model corresponding to a second stage decoder using said first type of frames, said reconstructed first type of frames, and a third representation by using a discriminator employed by a machine learning system, wherein said third representation is generated by compressing said reconstructed first type of frames by a second stage encoder;
• and
• generating reconstructed first and second types of frames by said generative model using soft edge information after training of said generative model, wherein said soft edge information is contained in a second representation generated by compressing said second type of frames of video by said second stage encoder.

A system, comprising:

• a memory for storing a computer program for compressing video;
• and
• a processor connected to the memory, wherein the processor is configured to execute the program instructions of the computer program comprising:
• partitioning a set of video frames into two subsets of different types of frames, a first type and a second type;
• compressing said first type of frames of videos to generate a first representation by a first stage encoder;
• and
• compressing said second type of frames of video to generate a second representation that only contains soft edge information by a second stage encoder, wherein said soft edge information comprises a soft map at a particular quantization level, wherein said soft map comprises a map of multilevel edges.

A system, comprising:

• a memory for storing a computer program for decompressing a set of video frames partitioned into two subsets of different types of frames, a first type and a second type;
• and
• a processor connected to the memory, wherein the processor is configured to execute the program instructions of the computer program comprising:
• decoding a first representation to reconstruct said first type of frames using a first stage decoder, wherein said first representation is generated by compressing said first type of frames of videos by a first stage encoder;
• training a generative model corresponding to a second stage decoder using said first type of frames, said reconstructed first type of frames, and a third representation by using a discriminator employed by a machine learning system, wherein said third representation is generated by compressing said reconstructed first type of frames by a second stage encoder;
• and
• generating reconstructed first and second types of frames by said generative model using soft edge information after training of said generative model, wherein said soft edge information is contained in a second representation generated by compressing said second type of frames of video by said second stage encoder.