Leveraged frame similarity to speed up CNN-based object detection and semantic segmentation models, e.g. FPN, PSPNet, YOLO, Faster-RCNN, and SegNet. Proposed a multimodal CNN achieving high fine-grained classification accuracy. The model was trained with both web images and its tags, and could predict solely with image in future testing phase. Under Graduate Research Assistant, SepAug 2014 Speech Processing Laboratory, NTU Researched in Deep Neural Network and Probabilistic Graphical Model, including Recurrent Neural Network and Regularized Mixture Model Surveyed various word embedding method, including Recurrent Neural Network (RNN), word2vec, paragraph vector.
國立台灣大學・
Electrical Engineering