I am trying to put together a list that in my opinion is essential to a computer vision engineer. As I am still working towards a higher tier of a engineer (see below for the definition what "rank" refers to here), I don't really know what really makes a CV engineer with Tier higher than 3. I will try to share it once I know.
(Chinese: 一個電腦視覺工程師的自我修養) This title was come from "An Actor Prepares". Because of Steven Chu's movie "[King of Comedy](https://en.wikipedia.org/wiki/King_of_Comedy_(film)", this book title become well-known in Chinese culture zone.
"Landau genius scale" or "Landau's ranking" [link] is a ranking system of physicists maintained by Lev Landau.
In "The Essence of computing" by Wu Jun (Chinese, currently there's no English version yet), Wu adopted similar concepts to ranking computer scientist or engineers by their impacts.
- Tier 1: Being able to create a whole new industry or lay out a foundation of an scientific already. Wu ranked Donald E. Knuth as this tier.
- Tier 2: Being able to make significant contributions to the theories or realize the products that other engineers cannot
- Tier 3: Being able to an unsolved problems and make successful engineering products in the markets
- Tier 4: Being able to lead a team to solve engineering problem using state-of-the art solutions.
- Tier 5: Being able to solve an engineering problem independently.
- Machine learning based classifier
- Regularization
- Stochastic Gradient Decent
- Concepts neural networks
- Fully-connected layers
- Convolutional network
- Backpropagation
- Optimizer
- Dropout
- Batch normalization
- Attention
- Well-known architectures: such as VGG or ResNet
- Image classification
- Fine-tuning
- Object detection
- Image Segmentation
- Deep Learning for Computer Vision, University of Michigan
- Deep Learning for Computer Vision, Stanford University
- Ian Goodfellow and Yoshua Bengio and Aaron Courville, Deep Learning
Dataset that is small enough and already hosted on Kaggle, so that it is possible to leverage on Kaggle's free GPUs.
- Image classification with Sports Image classification
- Car object detection with Car Object Detection dataset.
- Image Segmentation dataset:
- Coco Car Damage Detection Dataset
- Medical Image Segmentation: Evaluation: it seems interesting, but the scale is quite large
- Breast Ultrasound Images Dataset
- This scale of this dataset seems to be quite ideal (1578 files, 266.19 MB)
- Severstal: Steel Defect Detection
- 1.7GB data, a bit large
- Book: Software engineering at Google, online version, Purchase from O'Reilly