Bowen Cheng
AI Research Scientist at Meta
Menlo Park, California, United States
Summary
🤩
Rockstar🎓
Top SchoolBowen Cheng is an AI research scientist with nine years of experience building production-grade multimodal and computer vision systems, currently focused on multimodal personal intelligence at Meta after a stint as an MTS at OpenAI. He was a core contributor to Tesla Autopilot’s end-to-end FSD v12 and has a strong research foundation from a Ph.D. in ECE at UIUC under Alexander Schwing and Thomas Huang. Bowen’s work spans image recognition, segmentation, 3D occupancy, and multimodal perception—ship-ready systems informed by top-tier research. He is an active open-source contributor to influential projects like Detectron2 and Mask2Former, adding video instance segmentation and performance improvements used by the community. Bowen’s internships across FAIR, Google Research, and Microsoft Research reflect deep industry-academia collaboration and a knack for moving ideas from papers to production. Based in Menlo Park, he combines rigorous academic training with practical engineering that powers real-world perception and multimodal agents.
9 years of coding experience
9 years of employment as a software developer
Doctor of Philosophy - PhD Electrical and Computer Engineering, Doctor of Philosophy - PhD Electrical and Computer Engineering at University of Illinois Urbana-Champaign
Chinese, English
Github Skills (16)
mle10
semantic-segmentation10
computer-vision10
pytorch10
eval10
machine-learning10
instance-segmentation10
detectron10
python10
evaluation10
data-integration10
ml10
datasets10
data-set10
data-augmentation9
Programming languages (4)
Github contributions (5)
bowenc0221/panoptic-deeplab
Jun 2020 - Jan 2021
This is Pytorch re-implementation of our CVPR 2020 paper "Panoptic-DeepLab: A Simple, Strong, and Fast Baseline for Bottom-Up Panoptic Segmentation" (https://arxiv.org/abs/1911.10194)
Role in this project:
ML Engineer
ML EngineerContributions:79 commits, 24 PRs, 73 pushes in 7 months
Contributions summary:Bowen primarily contributed to the model's post-processing steps and evaluation metrics, adding support for different confidence score calculations for instance segmentation. They also added demo code for model inference and visualization. Furthermore, the user added support for the MobileNetV2 and Xception-65 backbones, which involved changes to configuration and potentially model architecture integration.
pytorchre-implementationarxivdetectron2abs
HRNet/HigherHRNet-Human-Pose-Estimation
Aug 2019 - Mar 2020
This is an official implementation of our CVPR 2020 paper "HigherHRNet: Scale-Aware Representation Learning for Bottom-Up Human Pose Estimation" (https://arxiv.org/abs/1908.10357)
Role in this project:ML Engineer
ML EngineerContributions:21 commits, 1 PR, 3 pushes in 6 months
Contributions summary:Bowen primarily contributed to the adaptation of the HigherHRNet model for the CrowdPose dataset. Their work included modifying the dataset loading and evaluation scripts to support CrowdPose, along with implementing necessary changes for multi-scale testing and flip-index configurations. They also made minor fixes to the dataset and evaluation procedures, ensuring proper functionality with the new dataset. Furthermore, the user made adjustments to the training setup, including SyncBN support and changes to the model configurations.
pytorchrepresentationhuman-posearxivabs
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.