1 paper
Yuanhuiyi Lyu, Xu Zheng, Jiazhou Zhou +1
We present UniBind, a flexible and efficient approach that learns a unified representation space for seven diverse modalities -- images, text, audio, point cloud, thermal, video, a…