You've learned a bit of OpenCV here, you took some Object Detection lessons there, maybe you learned some Deep Learning and are following the industry news... but still feels like it's not enough. It's because you may have been focusing on what's now seen as prerequisites, whereas...
To really move the needle, you should focus on the Computer Vision Ladder!
From: Jeremy Cohen
Dear friend, if you feel like randomly being pulled in all directions lately, and would like clarity on the exact Computer Vision skills to learn to make it into the field, then this page will show you a concrete path.
Here's the story:
This June, the annual Computer Vision and Pattern Recognition (CVPR) conference set the tone on many topics at the frontier of Computer Vision today: ideas like 3D Reconstruction, World Models, Vision Language Models, or Multi-Modal.
Chances are, most engineers reading this are FAR from that, focusing for weeks on learning skills that do not move the needle. Learning the core ideas behind object detection is interesting, but spending more time than needed on it doesn't grow your career.
So what does? In their CVPR 2026 talk, Intel defined a Ladder of Computer Vision that you can see on the right. It described 6 Levels that every engineer in Computer Vision should reach: Classification, Detection & Segmentation, Visual QA, 3D Vision, Video, and Reasoning.
- L6Reasoning
- L5Video
- L43D Vision
- L3Visual QA
- L2Detection & SegmentationMost Engineers never go past Level 2
- L1Classification
Most Engineers never go past Level 2.
If you paid attention, Level 2 is everything related to image processing: Image Segmentation, Object Detection, Pose Estimation, Keypoint detection, Oriented Bounding Boxes, and even... Tracking.
All of this, while interesting and necessary for every engineer, remains in the realm of Image Processing.
But today, Computer Vision is no longer about Image Processing
That was good for 2017. We evolved. I've interviewed over 20 field engineers working in the industry last year. Many were Computer Vision Engineers, some were CTOs, Director of Engineering of big groups, self-driving car engineers, autonomous robot engineers, even humanoid robot Engineers. Do you want to know what I learned from them?
Today's Computer Vision Engineers are very different from 2017's Computer Vision Engineers. Back then, you were asked to handle tools like PIL and OpenCV very well, you were asked to train YOLO on custom datasets, and to build segmentation masks. This part still exists, but has been swallowed by 3 core skills to understand:
Skill 1Deep Learning
Not just basic CNNs. This is beyond required. I also mean architectures (encoders, decoders, pyramidal blocks, ...) and Transformers (image, video, multi-modal ...)
Skill 23D Vision
Companies today are no longer afraid to go Physical. Understanding camera sensors, not images, understanding 3D, not 2D is what defines the new engineers.
Skill 3Video Perception
From series of frames to entire sequences... from tracking to reasoning and understanding, video is what gets you to the top of the ladder.
Understanding these skills is what I've been doing for the past decade. First, working as a Computer Vision Engineer for autonomous shuttles. I was building Deep Learning models for drivable area detection. Later on, working on 3D Computer Vision, Spatial AI, and Video Computer Vision.
I have gathered what I learned for over a decade in a series of 6 courses that put together form THE COMPUTER VISION PACK.