Benign Innovations
Case study 
Arqam ImtiazMachine Learning Engineer · Shenzhen, China
Computer Vision · Robotics · Edge AI
I build perception systems for the real world.
From dataset strategy and auto-labeling to model optimization and edge deployment, I turn computer-vision research into dependable product systems.
7+ yearsComputer vision and robotics engineering
Data → edgeFull model-lifecycle ownership
Real timeTemporal, multi-camera, and embedded systems
Perception lifecycleProduction-minded
PerceptionProduct system
01 · DataCurate · label · audit
02 · ModelTrain · evaluate · iterate
03 · EdgeOptimize · deploy · profile
04 · LearnMonitor · mine · improve
Proof in motion
Watch the systems work.
Real deployments, not presentation mockups. Play a demo here, then open the case study for the engineering behind it.
Benign Innovations
Case study Auto-Labeled Weed Detection and Species Classification
LinxAI Intelligent Technology
Case study Deep Reinforcement Learning for Custom Quadruped Locomotion
Selected work
Systems, not just models.
A focused selection of computer-vision and robotics work spanning data design, model development, deployment, and real-world operation.
Dual-Camera 3D Tennis Scene Understanding
A multi-model perception system that maps court geometry, players, and fast-moving ball trajectories into real-world 3D coordinates.
Real-Time Tennis Analytics with Multi-Modal Vision on Edge DevicesIncludes video
A real-time, edge-powered tennis analytics stack combining player tracking, pose, ball trajectory, and court mapping.
Auto-Labeled Weed Detection and Species ClassificationIncludes video
A lawn-care perception component that distills foundation-model masks and classifications into a compact weed detector.
Deep Reinforcement Learning for Custom Quadruped LocomotionIncludes videoIncludes image
Proprioception-only locomotion policies transferred from simulation to a custom quadruped across flat and rough terrain.
Instance Segmentation from Auto-Labeling to Edge DeploymentIncludes video
An end-to-end instance-segmentation workflow spanning foundation-model auto-labeling, training, evaluation, and edge deployment.
Dense Depth Labels from LiDAR–Camera Fusion
A synchronized LiDAR and multi-view pipeline that increased depth-label density from 6% to 58% through calibrated frame fusion.
Experience
Building where models meet products.
My work has moved between sports AI, autonomous robots, 3D perception, and multi-camera systems—always with deployment constraints in view.
Enhanced Robotics (Tenniix Official)
Algorithm Engineer
Own the visual-perception lifecycle for real-time tennis AI systems.
- Auto-labeling and dataset iteration
- Detection, tracking, and court geometry
- Edge deployment and monitoring
Benign Innovations (Co-founding Startup)
Algorithm Engineer
Led applied perception R&D for tennis analytics and autonomous lawn-care robotics.
- Multi-modal tennis analytics
- Robot perception and docking
- Practical edge-model deployment
LinxAI Intelligent Technology Co., Ltd.
Algorithm Engineer
Developed deep-reinforcement-learning locomotion policies and simulation-to-robot transfer workflows.
- Quadruped locomotion
- Isaac Gym simulation
- Robust terrain control
XPENG Robotics
Deep Learning Engineer
Built and optimized perception workflows across indoor 3D vision, robot interaction, and autonomous systems.
- 3D and stereo perception
- Human-centric perception
- Model development and optimization
Southern University of Science and Technology
Research Assistant
Researched 3D scene parsing across semantic mesh segmentation and point-cloud object detection.
- 3D geometric analysis
- Mesh data preparation
- LiDAR scene understanding
ROPEOK Technology Group
Algorithm Engineer
Developed real-time person re-identification and attribute-recognition systems for multi-camera environments.
- Person re-identification
- Human attributes
- Multi-camera deployment
Pakistan Aeronautical Complex
Design Engineer
Contributed to smart-display graphics and EEG feature-extraction initiatives.
- OpenGL display software
- Signal feature extraction
- Applied engineering R&D
Capabilities
Across the full ML lifecycle.
I’m most useful on problems where data quality, model behavior, hardware limits, and product requirements need to be solved together.
Production vision
Real-time detection, tracking, temporal models, keypoints, segmentation, and multi-camera geometry.
- Detection & tracking
- Temporal vision
- Multi-camera systems
Data flywheels
Auto-labeling, active learning, hard-example mining, synthetic data, and quality auditing.
- Auto-labeling
- Dataset curation
- Quality loops
Model to edge
Training, benchmarking, ONNX and TensorRT export, quantization, profiling, monitoring, and iteration.
- PyTorch
- ONNX / TensorRT
- Performance profiling
Robotics & 3D
Depth, pose, point clouds, camera calibration, simulation-to-real transfer, and spatial perception.
- 3D perception
- Robot learning
- Sensor geometry
Core toolchainPythonC/C++PyTorchOpenCVTensorRTONNXDockerLinuxMLflow
Education
The engineering foundation.
Xiamen University
Master of Engineering · Computer Technology
Thesis: Deep discriminative feature learning with a multi-level network for person re-identification.
COMSATS University
Bachelor of Science · Electrical Engineering
Final project: vision-based shooting-target accuracy measurement for a training range.
Start a conversation
Working on a difficult perception problem?
Interested in production computer vision, robotics perception, sports AI, and edge ML work.