A 12-week intensive for engineers who are tired of copying PyTorch snippets they don't understand. Ship a production-grade object detection pipeline. Deploy it on edge hardware. Know exactly why it works.
epoch 47 / loss
0.0312 ↓
mAP@0.5
97.3%
0.0%
Best project mAP
0
Shipped projects per cohort
0+
Engineers trained
0ms
Fastest edge deployment
Every card is a real system a student built and deployed. Hover to see the architecture, dataset, and accuracy.
Catching micro-cracks at 120 fps on a conveyor belt
Marcus Webb
Senior SWE, Bosch
Real-time depth estimation and path planning at 30m/s
Priya Nair
Robotics Lead, Zipline
Grading diabetic retinopathy from fundus photographs
Dr. James Okafor
ML Eng, PathAI
12 shipped projects per cohort · All source code available to graduates
Each module is reverse-engineered from a real shipping requirement. No filler lectures.
Build intuition for image representation — color spaces, convolution from scratch, and why 3×3 kernels matter more than you think.
Dissect ResNet, VGG, and EfficientNet layer by layer. Visualize activations. Build a feature extractor you actually understand.
Anchor boxes, IoU, NMS — the primitives behind every production detector. Implement YOLO's loss function before you use it.
YOLOv8, EfficientDet, DETR. When to use which. Fine-tune on custom data. Avoid the annotation traps that kill real projects.
Monocular depth estimation, stereo vision, and point cloud processing. Build the spatial understanding robotics teams pay for.
ONNX export, TensorRT optimization, Jetson Nano deployment. Shrink a 100ms model to 8ms without destroying accuracy.
14
Papers Published
800+
Students Trained
9
Industry Years
Former Research Scientist, NVIDIA Research
Elena spent seven years building perception stacks for autonomous vehicles at NVIDIA before pivoting to teaching. Her course at CMU's MCDS program ran three cohorts before she launched Perceive. She ships code before she writes slides.
Selected Publications
Efficient Multi-Scale Feature Fusion for Real-Time Object Detection
CVPR · 2022
Quantization-Aware Training for Edge Deployment of Detection Networks
ICCV · 2023
8
Papers Published
23
Models Deployed
11
Industry Years
Principal ML Engineer, Waymo (prev. Google Brain)
Rahul designed the perception evaluation framework used by Waymo's sensor fusion team. He believes the gap between academic CV and production CV is a pedagogy problem, not a research problem — and built this curriculum to close it.
Selected Publications
Scalable Annotation Pipelines for Autonomous Driving Datasets
NeurIPS Workshop · 2021
Cross-Domain Adaptation for Adverse Weather Perception
ECCV · 2022
4 hours of async lecture, 2 hours live Q&A, 2 hours project work. Designed around an engineering schedule.
One shipped project per week. Graded on accuracy, latency, and code quality — not just completion.
Jetson Nano kit included. You deploy on real hardware — not a Colab notebook.
Small enough for direct instructor access. Large enough for peer review that actually helps.
March 2026 cohort. 40 seats. Application is two questions — no resume, no interview.
48 pages. Every lecture topic, every paper on the reading list, every project rubric. If you want to see exactly what 12 weeks covers before committing, this is it.