Experience

Flexiv Ltd.

• United Kingdom
• Facilities Services

• Senior Computer Vision Algorithm Engineer

  • Jul 2019 - Present

  1. Conducted research and implementation on SOTA computer vision algorithms, such as object detection (include 3D and oriented object), instance segmentation, etc. 2. Designed and helped build the Noema training cloud platform, a complete pipeline from data labeling, manipulation and augmentation, to module training and deployment. 3. Developed a highly configurable training framework, with pluginable tools like data augmentation and model architecture customization, model compression and optimization, etc. 4. Created an inference engine featuring easy-to-design multi-model workflows. Show less



Baidu, Inc.

• China
• Software Development
• 700 & Above Employee

• Computer Vision Engineer

  • Nov 2018 - Jul 2019

  1. Facial attribute editing - Utilize face landmark detection and Poisson matting, to change the face in the original video. - The success rate is up to 80% for age swapping and 71% for gender swapping. 2. PaddleGAN Open-source development Develop representative GAN models using Paddle, including pix2pix, attGAN and stGAN. 3. Video understanding and audit - Use Flownet-simple to detect shaking videos, precision and recall are up to 90%/60%, used for video searching in Baidu, greatly improved searching experience. - In charge of the anti-porn and anti-terror model optimization and iteration. Show less



Huawei

• Australia
• Renewable Energy Semiconductor Manufacturing
• 1 - 100 Employee

• Computer Vision Engineer Intern

  • Apr 2018 - Oct 2018

  Research on 3D reconstruction and depth estimation 1. Depth completion: Predict the surface normal and boundary, and then combine the raw depth image to get the complete depth image. 2. Transplant code from torch to caffe framework, run on Kirin 980's npu. Research on 3D reconstruction and depth estimation 1. Depth completion: Predict the surface normal and boundary, and then combine the raw depth image to get the complete depth image. 2. Transplant code from torch to caffe framework, run on Kirin 980's npu.



DJI

• China
• Consumer Electronics
• 700 & Above Employee

• Computer Vision Research Engineer Intern

  • Aug 2017 - Oct 2017

  Research on image Super Resolution - Use the Laplacian layer structure, utilize two parallel networks to extract the image feature and reconstruct the image simultaneously. - Add two residual feedback, increase the PSNR by 0.1 dB. Research on image Super Resolution - Use the Laplacian layer structure, utilize two parallel networks to extract the image feature and reconstruct the image simultaneously. - Add two residual feedback, increase the PSNR by 0.1 dB.