</> Code with Papers by Syntology

Papers › SPICE: Semantic Pseudo-labeling for Image Clustering

SPICE: Semantic Pseudo-labeling for Image Clustering

17 Mar 2021arXiv:2103.09382archive 2025-07-28

Chuang Niu, Hongming Shan, Ge Wang

The similarity among samples and the discrepancy between clusters are two crucial aspects of image clustering. However, current deep clustering methods suffer from the inaccurate estimation of either feature similarity or semantic discrepancy. In this paper, we present a Semantic Pseudo-labeling-based Image ClustEring (SPICE) framework, which divides the clustering network into a feature model for measuring the instance-level similarity and a clustering head for identifying the cluster-level discrepancy. We design two semantics-aware pseudo-labeling algorithms, prototype pseudo-labeling, and reliable pseudo-labeling, which enable accurate and reliable self-supervision over clustering. Without using any ground-truth label, we optimize the clustering network in three stages: 1) train the feature model through contrastive learning to measure the instance similarity, 2) train the clustering head with the prototype pseudo-labeling algorithm to identify cluster semantics, and 3) jointly train the feature model and clustering head with the reliable pseudo-labeling algorithm to improve the clustering performance. Extensive experimental results demonstrate that SPICE achieves significant improvements (~10%) over existing methods and establishes the new state-of-the-art clustering results on six image benchmark datasets in terms of three popular metrics. Importantly, SPICE significantly reduces the gap between unsupervised and fully-supervised classification; e.g., there is only a 2% (91.8% vs 93.8%) accuracy difference on CIFAR-10. Our code has been made publically available at https://github.com/niuchuangnn/SPICE.

PaperPDFConference PDFCode

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

niuchuangnn/SPICE officialmentioned in papermentioned on GitHubpytorch report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

ClusteringContrastive LearningDeep ClusteringImage ClusteringRepresentation LearningSemantic SimilaritySemantic Textual Similarity

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Image Clustering CIFAR-10 SPICE* ARI 0.836 #9 of 40 Archive leaderboard report
Image Clustering CIFAR-10 SPICE* Accuracy 0.918 #9 of 40 Archive leaderboard report
Image Clustering CIFAR-10 SPICE* Backbone ResNet-18 #9 of 40 Archive leaderboard report
Image Clustering CIFAR-10 SPICE* NMI 0.850 #9 of 40 Archive leaderboard report
Image Clustering CIFAR-10 SPICE* Train set Train #9 of 40 Archive leaderboard report
Image Clustering CIFAR-100 SPICE* ARI 0.422 #7 of 30 Archive leaderboard report
Image Clustering CIFAR-100 SPICE* Accuracy 0.584 #7 of 30 Archive leaderboard report
Image Clustering CIFAR-100 SPICE* NMI 0.583 #7 of 30 Archive leaderboard report
Image Clustering CIFAR-100 SPICE* Train Set Train #7 of 30 Archive leaderboard report
Image Clustering ImageNet-10 SPICE ARI 0.933 #2 of 18 Archive leaderboard report
Image Clustering ImageNet-10 SPICE Accuracy 0.969 #2 of 18 Archive leaderboard report
Image Clustering ImageNet-10 SPICE Backbone ResNet-34 #2 of 18 Archive leaderboard report
Image Clustering ImageNet-10 SPICE NMI 0.927 #2 of 18 Archive leaderboard report
Image Clustering Imagenet-dog-15 SPICE ARI 0.526 #8 of 20 Archive leaderboard report
Image Clustering Imagenet-dog-15 SPICE Accuracy 0.675 #8 of 20 Archive leaderboard report
Image Clustering Imagenet-dog-15 SPICE Backbone ResNet-34 #8 of 20 Archive leaderboard report
Image Clustering Imagenet-dog-15 SPICE NMI 0.627 #8 of 20 Archive leaderboard report
Image Clustering STL-10 SPICE* Accuracy 0.929 #6 of 29 Archive leaderboard report
Image Clustering STL-10 SPICE* Backbone ResNet-34 #6 of 29 Archive leaderboard report
Image Clustering STL-10 SPICE* NMI 0.860 #6 of 29 Archive leaderboard report
Image Clustering STL-10 SPICE* Train Split Train #6 of 29 Archive leaderboard report
Image Clustering Tiny-ImageNet SPICE ARI 0.161 #3 of 14 Archive leaderboard report
Image Clustering Tiny-ImageNet SPICE Accuracy 0.305 #3 of 14 Archive leaderboard report
Image Clustering Tiny-ImageNet SPICE NMI 0.449 #3 of 14 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

Contrastive Learning

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections