Xinyi Li Email & Phone Number

Cambridge, MA, US

Boston, Massachusetts, United States

• Established a framework of optical flow estimation for training, evaluation and prediction by using PyTorch, OpenCV and NumPy, to trace the motion of objects between successive frames.
• Designed a comprehensive dataset to evaluate the performance of optical flow models on various data with varying degrees of noise, motion, visual effects, etc., and wrote a data loader to pre-process data in various.
• Implemented the state-of-the-art optical flow estimation deep learning algorithms and performance metrics in the team’s internal pipeline, and reduced the prediction errors by adjusting the models, data and training.
• Applied various optical flow models into other application pipelines, such as burst denoise and video frame interpolation, and improved their generated results slightly.

• Designed a Multi-block Attention module, a Multi-block Feature Fusion module and a new Graph Routing method to improve the robustness and classification performance of the capsule network with much fewer parameters.
• Completed the paper named Global Fusion Capsule Network with Pairwise-Relation Attention Graph Routing as the first author, which was accepted by ICONIP(International Conference on Neural Information Processing).
• Conducted abundant experiments of object classification and person re-identification with PyTorch and TensorFlow.

• Assisted my team to construct an OCR (Optical Character Recognition) model to identify train tickets by combining OCR modules for the recognition of English, numeric, and Chinese characters.
• Mainly responsible for reviewing state-of-the-art papers and models, generating experimental data, and conducting massive experiments with PyTorch to establish the most effective models.

Master Of Science - Ms, Computer Science

Bachelor Of Engineering - Be, Automation, 92.14/100