CliqueCNN: Deep Unsupervised Exemplar Learning

By Miguel Angel Bautista, Artsiom Sanakoyeu, Ekaterina Tikhoncheva, Björn Ommer. 2016

116 citations recorded by the source corpus.

Exemplar learning is a powerful paradigm for discovering visual similarities in an unsupervised manner. In this context, however, the recent breakthrough in deep learning could not yet unfold its full potential. With only a single positive sample, a great imbalance between one positive and many negatives, and unreliable relationships between most samples, training of Convolutional Neural networks is impaired. Given weak estimates of local distance we propose a single optimization problem to extract batches of samples with mutually consistent relations. Conflicting relations are distributed over different batches and similar samples are grouped into compact cliques. Learning exemplar similarities is framed as a sequence of clique categorization tasks. The CNN then consolidates transitivity relations within and between cliques and learns a single representation for all samples without the need for labels. The proposed unsupervised approach has shown competitive performance on detailed posture analysis and object classification.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Invariant Information Clustering for Unsupervised Image Classification and Segmentation (2018, methodology): “Agglomerative clustering [2] and partially ordered sets [1] of HOG features [13] have been used to group images, and exemplars [18] define a group as a set of random transformations applied to a single image.”
Deep Clustering for Unsupervised Learning of Visual Features (2018, background): “Other clustering losses [3, 16, 35, 66, 68] have been considered to jointly learn convnet features and image clusters but they have never been tested on a scale to allow a thorough study on modern convnet architectures.”
Unsupervised Learning of Visual Features by Contrasting Cluster Assignments (2020, neutral): “Our work is also related to clustering-based methods [2, 3, 6, 7, 17, 25, 50, 53, 54, 59].”
Self-labelling via simultaneous clustering and representation learning (2019, neutral): “, 2017), clustering (Caron et al., 2018; Huang et al., 2019; Zhuang et al., 2019; Bautista et al., 2016), noise-as-targets (Bojanowski & Joulin, 2017), colorization (Zhang et al.”
Cross and Learn: Cross-Modal Self-Supervision (2018, background): “There is also a variety of works which specifically aim to learn reliable posture representations without supervision [3,30,23,32,4].”
VICReg: Variance-Invariance-Covariance Regularization for Self-Supervised Learning (2021, methodology): “Instead of viewing each sample as its own class, clustering-based methods [3, 46, 44, 27, 51, 6, 1, 45, 18, 5, 7] regroup them into clusters based on some similarity measure.”
Deep Comprehensive Correlation Mining for Image Clustering (2019, background): “Instead of clustering, several approaches [3, 25, 34, 13, 39, 2, 47, 49] mainly focus on deep unsupervised learning of representations.”
A Variational U-Net for Conditional Appearance and Shape Generation (2018, result): “In contrast to our generative approach, [4, 3] have pursued unsupervised learning of human posture similarity for retrieval in still images and [25, 5] in videos.”
Neighborhood Contrastive Learning for Novel Class Discovery (2021, result): “To this end, many semi-supervised [5, 35] and unsupervised learning [4, 7, 14, 32] methods have been proposed, which achieve promising results compared to supervised learning methods.”
A Bottom-Up Clustering Approach to Unsupervised Person Re-Identification (2019, background): “A number of works (Dosovitskiy et al. 2014; Bautista et al. 2016) sample patches from images and generate labels for the patches as supervision.”
A Style-Aware Content Loss for Real-time HD Style Transfer (2018, background): “Other works like [48,5,34,9,1] learn how to discriminate different techniques, styles and contents in the latent space.”
Video Jigsaw: Unsupervised Learning of Spatiotemporal Context for Video Action Recognition (2018, methodology): “Other unsupervised representation learning approaches include exemplar CNNs [11], CliqueCNNs [4] and unsupervised similarity learning by clustering [3].”
Leveraging Large-Scale Uncurated Data for Unsupervised Pre-training of Visual Features (2019, background): “Various attempts have been made to train convnets using clustering [2, 3, 6, 29, 48, 52, 53].”
Rethinking Semantic Segmentation: A Prototype View (2022, background): “Since computing all the pairwise comparisons on a large dataset is challenging, some clustering-based methods turn to discriminate between groups of images with similar features instead of individual images [2, 5, 12, 13, 64, 104, 119, 121, 122].”
Class-Incremental Few-Shot Object Detection (2021, methodology): “Moreover, although we use the simple but effective clustering method (k-means) in this paper, other exemplar learning methods are worth trying in the future [Bautista et al., 2016; Mairal et al., 2008].”
Efficient Training of Visual Transformers with Small Datasets (2021, background): “Most current self-supervised computer vision approaches can be categorised in contrastive learning [54, 25, 9, 23, 52, 56, 18], clustering methods [3, 67, 29, 5, 1, 20, 6, 7], asymmetric networks [22, 10] and feature-decorrelation methods [19, 63, 2, 27].”
Broaden Your Views for Self-Supervised Video Learning (2021, unknown): “Clustering-based methods [5, 8, 11, 13, 14, 38, 77, 84] alternate between learning representations using clusters as targets, and clustering using the current representations (either online or offline).”
Semi-Supervised Segmentation of Salt Bodies in Seismic Images using an Ensemble of Convolutional Neural Networks (2019, background): “k of the annotations seismic image interpretation did not allow to unfold the full potential of the CNNs. The recent trend in the Computer Vision community is unsupervised or self-supervised learning [9,33,34,24,2, 1,39,5,19,28] which can make use of an abundant unlabeled visual data available on th”
With a Little Help from My Friends: Nearest-Neighbor Contrastive Learning of Visual Representations (2021, background): “Methods that use prototypes/clusters [1, 2, 5, 6, 7, 23, 33, 53, 56, 57, 58, 60] also attempt to learn features by associating multiple samples with the same cluster.”
Deep Unsupervised Similarity Learning Using Partially Ordered Sets (2017, methodology): “To evaluate the valid- 3https://github.com/asanakoy/deeppose_tf ity of our representation we compare the performance obtained by DeepPose [29], when trained with one of the following models as initialization: random initialization, Shuffle&Learn [20] (triplet model), and our approach trained on OS.”