Image Classification
Assigning one label to a whole image, such as cat, dog or car. The task that kicked off the deep learning era.
Also image recognition · ImageNet
Show a picture and ask: what is this? The computer answers with one word, and how sure it is.
Where it fits
+All relationships as a list
Is a kind of / part of
Siblings
Uses / built on
Used by
How it works
An image classifier takes pixels in and produces a probability for each class out. Training is ordinary supervised learning: show millions of labeled images, compare the predicted probabilities with the true label using cross-entropy, and update the network with backpropagation.
The ImageNet challenge (1.2 million training images, 1,000 classes) made classification the yardstick of progress. In 2012 AlexNet cut the error rate dramatically with a CNN; within a few years, networks like ResNet beat a commonly cited estimate of human error on that benchmark (about 5% top-5 error).
Classification is also the starting point for harder tasks: the same trained network, minus its last layer, becomes the “backbone” that detection and segmentation models build on.
Everyday use cases
- Case 01
Plant and bird identification apps
Point your phone at a flower and get its species.
- Case 02
Quality control
Sort products on a conveyor belt into good and defective.
Famous examples
- 2012AlexNet
The deep convolutional network that won the 2012 ImageNet challenge by a huge margin and kicked off the deep learning boom.
- 2015Photo Library Search
Typing 'dog', 'beach' or 'receipt' into your photo app and finding matching pictures you never tagged, powered by image classifiers and image embeddings.
Strengths, weaknesses and when to use it
Strengths
- Simple to set up and evaluate (accuracy, top-5 accuracy).
- Pretrained classifiers transfer well to new tasks.
Weaknesses
- One label per image; it cannot say where the object is or count several.
- Can rely on shortcuts, such as backgrounds, instead of the object itself.
When to use itWhen each image has one main subject and you only need to know what it is.
Compare with
Check yourself
- Q01
What does an image classifier output?
Show answer
B: A label (with probabilities) for the whole image