Computer vision series: Introduction to CNN for Visual Recognition

History

Benchmark Datasets

• Pascal VOC
• Imagenet : 22k categories, 14M images

Imagenet Large scale visual recognition challenge 1000 Object Classes, Large number of images 1.4M images in an experiment error rate has decreased a lot. That's a lot of progress made

In 2012, error rate decreased significantly due to CNN (Convolutional Neural Network) a deep learning

Most important problem of Visual recognition - image classification

In image classification - your algorithim looks at an image, and then picks from among some fixed set of categories to classify the image. This relatively basic tool of image classification is super useful and can be applied all over the place for many different applications.

There is a number of visual recognition problems that are related to image classification, such as object detection, image captioning.

The setup in object detection is a little bit different. Rather than classifying an entire image as cat or dog or whatsnot. Instead we want to go in and draw bounding boxes and say there is a dog here, and a cat here and a car over in the background and draw these boxes describing where objects are in the image.

Image Captioning: ...

CNNs aka ConvNets

CNN have become an important tool for object recognition. learn more about CNN (Deep dive)

Layers of network. AlexNet - 7 to 8 layer 2014 GoogleNet VGG - 19 layers

The more layer, the better. i.e 200 layers but you will run out of GPU (discuss later)

**Main takeway: ** CNN really had this breakthrough moment in 2012, since then there's been a lot of effort focued in tuning and tweaking this algorithims to make them perform better and better on this problem of image classification. We will dive deep exactly how these different models work.

CNNs were not invented overnight. Why it's popular in recent Years? Faster and faster computers, GPUs (Thanks to moore's law). It allowed researcher to explore larger architecture and models. It's super important in the history of CNN.

Other keywords: Style Transfer