CVPR 2022 - learning non target knowledge for semantic segmentation of small samples

Due to the full convolution network （Fully Convolutional Network, FCN） The rapid development of architecture , Deep learning has made milestone progress in semantic segmentation . Most methods use a fully supervised learning program , Need a lot of annotated data for training . Although they can achieve good performance , But their data hungry nature requires a lot of pixel level image annotation .

To alleviate this problem , Dr. Shaoling, chief scientist of Tesla and his team , A framework for semantic segmentation of small samples is proposed , A support set for annotation at a given number of pixel levels （Support） In the case of images , Split query set （Query） The target object in the image . Relevant research results have been made in 2022 year CVPR publish , Titled 《 Learn non target knowledge for semantic segmentation of small samples 》（Learning Non-target Knowledge for Few-shot Semantic Segmentation）.

Thesis link ：https://arxiv.org/pdf/2205.04903.pdf