Research Project Topic: Learning To Play Hide-and-Seek
Author: Daming Wang
Supervisor: Dr.Michael Burke
The proposed approach is an image temporal ranking approach of learning the latent interest's information from demonstration video sequences. Also, the automatic labelling method based on the learning from time idea is investigated in this research project. It is proved that learning from time can satisfy this unsupervised learning task and replace manual labelling.
Further, the ranking models have experimented on different scenarios test videos. The benefits of the probabilistic approach are investigated in natural images, which make the model more conservative than deterministic models.
In addition, to improve the ranking models performance, the transfer learning study takes different convolutional neural network models for the feature extraction to discovers the better options.
More Detail in Uploaded Poster...