Third prize in the Zero-shot segmentation of Zhijiang Cup Global Artificial Intelligence Competition (2020).
Date:
Xing Tao, Zifan Chen
We have won third prize in the Zero-shot segmentation of Zhijiang Cup Global Artificial Intelligence Competition (2020). This competition aims to train a model that can be used to segment the object of unknown categories in images. Our solution divides the pipeline into two phases, class-independent main object segmentation and object-based zero-shot classification. We first use BERT to extract the word vectors of categories to establish the relationship between unknown categories and known categories. Then we can map the probability of known categories to the distribution of unknown categories according to the relationship. The overall solution is simple, and we believe it still has much room for improvement.