- Name: Xiaofeng Zhang
- Email: [email protected]
- Phone: (+86) 15392924328
- Degree: Master of Information Technology
- Period: 2021 - 2022
- Degree: Bachelor of Electronic & Computer Engineering
- Period: 2019 - 2020
- Degree: Bachelor of Electronic & Computer Engineering
- Period: 2019 - 2020
- Position: Research Scientist
- Period: 2022 - Present
- Development Perception model for Grount Truth System: Based on PersFormer and BEVFusion, use Lidar point as 3D PE to guide the height of BEV feature map. Apply min-cost flow for matching to contrast DETR-like loss.
- Data selection: use Perceptre-hash, Timestamps, and IMU information to select divergence data for human annotation.
- Annotation quality check: use CLIP-IQA to select blurred, bad samples from the dataset. Using Grounded-SAM and other rule-based functions to check lane line labeling quality.
- Build up a local registry for docker images.
- Fulfill ITL network requirements & DevOps by using Ansible and playbook.
- Build a monitor based on Grafana & Prometheus.
- Build high-concurrent DFS for computer vision case by Ceph.
- Position: Research Assitant
- Period：2019-2021
- Unity Client Development: Build a digital twin for Toyota Infotech based on Unity Engine. Adapt LGSVL to the original project, and build the communication among SUMO's traffic flow, real-world vehicles, and Unity Engine.
- Tech Art: Optimize graphic performance for PBR rendering to ensure the Oculus Rift2 can run at least 45Hz on RTX2080.
- Use VGG's muti-scale contrast an image vector as key for image data, and use cosine distance to search related images in the database, use OANet to calculate the query image's orientation & position.
- Adapt a trained on GAT5's DeepLabV3 model to the Citysapes dataset via labels from semantic-kitty.