Lecture 2 | Image Classification

#Image Classification : core task in Computer Vision

-the problem : Semantic Gap

-chanllenges

: Viewpoint variation - all pixels change when the camera moves

: Illumination - 명암

: Deformation - 다양한 포즈

: Occlusion - 부분만 보이는 경우

: Background Clutter - 배경

: Intraclass variation - 여러 종류의 형태로 존재

#An Image Classifier

-Data-Driven Approach

1. Collect a dataset of images and labels

2. Use Machine Learning to train a classifier

3. Evaluate the classifier on new images

#Nearest Neighbor

-train : Memorize all data and labels

-predict : Predict the label of the most similar training image

#K-Nearest Neighbor 

- Instead of copying label from nearest neighbor, take majority vote form K closest points.

- CIFAR10

- fast train, slow test 

-Distance Metric

: L1(Manhattan) distance

: L2(Euclidean) distance

-Hyperparameter : Distance metric and K

:Cross-Validation : Split data into folds, try each fold as validation and average the results

                       : Useful for small datasets, but not used too frequently in deep learning

-k-Nearest Neighbor on images never used.

:Very slow at test time

: Distance metrics on pixels are not informative

: Curse of dimensionality

#Linear Classification

-(one of )Parametric model

-f(x, W) : x = input data, W = parameters or weights -> return 10 class scores

-bias : constant vector of 10 elements that does not interact with the training and instead just gives us some data independent preferences for some classes over another. 

       : if dataset is unbalanced, and has many more cats than dogs for example, then the bias element corresponding to cat will be higher than the other ones.

-Hard cases for a linear classifier

: parity problem - 1과 0으로 성립된 수열에 있어서, 1의 개수의 짝,홀을 나타내는 말. 그 개수가 짝수일 때 패리티는 0, 홀수일 때 패리티는 1이라고 함.

: multi-model situation

Assignment 1

-K-Nearest Neighbor

-Linear classifiers: SVM, Softmax

-Two-layer neural network

-Image features