Image processing method and apparatus based on super network, and computer storage medium

An image processing apparatus based on a super network, comprising:

• a processor;
• and
• a memory that is configured to store instructions executable for the processor,
• wherein the processor is configured to execute the instructions to:
• merge a pretrained backbone network with a rear end of a target detection network to obtain a merged super network, the pretrained backbone network including a classified super network in a search space of the backbone network, the classified super network being configured to extract an image feature from image information, and the rear end of the target detection network including a target detection rear-end super network in a search space of the rear end of the target detection network, the target detection rear-end super network being configured to perform target detection based on the image feature;
• train the merged super network;
• perform Neural Architecture Search (NAS) based on the trained super network to obtain a target detection neural architecture;
• and
• process an image to be processed using the target detection neural architecture to obtain an image processing result.

A non-transitory computer-readable storage medium having computer programs stored thereon that, when executed by a processor, cause the processor to implement operations comprising:

• merging a pretrained backbone network with a rear end of a target detection network to obtain a merged super network, the pretrained backbone network including a classified super network in a search space of the backbone network, the classified super network being configured to extract an image feature from image information, and the rear end of the target detection network including a target detection rear-end super network in a search space of the rear end of the target detection network, the target detection rear-end super network being configured to perform target detection based on the image feature;
• training the merged super network;
• performing Neural Architecture Search (NAS) based on the trained super network to obtain a target detection neural architecture;
• and
• processing an image to be processed by using the target detection neural architecture to obtain an image processing result.