Projects
My research topics are computer vision and robotics. The followings are the project.
Instance Segmentation by Semi-Supervised Learning and Image Synthesis
Instance Segmentation is the method to detect objects' location and class. Since instance segmentation requires huge annotation costs, it is difficult to get a lot of data. To solve this problem, we propose the method to create a lot of annotated data by combining semi-supervised learning and image synthesis. For more detail, see this page.
Reducing error accumulation through recursive motion and video prediction
Imitation learning is a key technique for planning the robot motion. One of the major problems of supervised imitation learning is accumulation error. To solve this problem, we propose to use video prediction to reduce the accumulation error with some technical tricks. We will add the link to this project paper link later.
Diffusion + Retrieval
Current Project!!
Diffusion model is one of the state-of-the-art generative models, but it suffers from issues such as inference speed and generating inappropriate motions depending on the initial noise. Therefore, by utilizing retrieval, we can enhance the diffusion model by obtaining appropriate initial values.