Abstract:We devise deep nearest centroids (DNC), a conceptually elegant yet surprisingly effective network for large-scale visual recognition, by revisiting Nearest Centroids, one of the most classic and simple classifiers. Current deep models learn the classifier in a fully parametric manner, ignoring the latent data structure and lacking simplicity and explainability. DNC instead conducts nonparametric, case-based reasoning; it utilizes sub-centroids of training samples to describe class distributions and clearly explains the classification as the proximity of test data and the class sub-centroids in the feature space. Due to the distance-based nature, the network output dimensionality is flexible, and all the learnable parameters are only for data embedding. That means all the knowledge learnt for ImageNet classification can be completely transferred for pixel recognition learning, under the "pre-training and fine-tuning" paradigm. Apart from its nested simplicity and intuitive decision-making mechanism, DNC can even possess ad-hoc explainability when the sub-centroids are selected as actual training images that humans can view and inspect. Compared with parametric counterparts, DNC performs better on image classification (CIFAR-10, ImageNet) and greatly boots pixel recognition (ADE20K, Cityscapes), with improved transparency and fewer learnable parameters, using various network architectures (ResNet, Swin) and segmentation models (FCN, DeepLabV3, Swin). We feel this work brings fundamental insights into related fields.

A Pseudo Nearest Centroid Neighbour Classifier

A multi-average based pseudo nearest neighbor classifier

A pre-averaged pseudo nearest neighbor classifier

A Novel Pseudo Nearest Neighbor Classification Method Using Local Harmonic Mean Distance

The Nearest Neighbor Algorithm of Local Probability Centers

K-Ns: A Classifier by the Distance to the Nearest Subspace.

A Bayesian Network nearest k-labels method for Multi-label classification

PCNN: Probable-Class Nearest-Neighbor Explanations Improve Fine-Grained Image Classification Accuracy for AIs and Humans

Nearest neighbor classification using cam weighted distance

Active and Passive Nearest Neighbor Algorithm: A Newly-Developed Supervised Classifier

Improving Nearest Neighbor Classification with Cam Weighted Distance.

Nearest Feature Line Classifier Based on Collaborative Representation with Nearest Neighbour Search Algorithm

In defense of Nearest-Neighbor based image classification

Local Naive Bayes Nearest Neighbor for Image Classification

Tunable Nearest Neighbor Classifier

A Fast Document Classification Algorithm Based on Improved KNN

Visual Recognition with Deep Nearest Centroids

K Nearest Neighbor Based Local Sparse Representation Classifier

Combining KNN algorithm and other classifiers.

Theoretical Analysis of the Confidence Metrics for Nearest Neighbor Classifier

A Novel Nearest Neighbors Algorithm Based on Power Muirhead Mean