Handle the problem of ample label space by using the Image-guided Feature Extractor on the MUSTI dataset