Papers › Improving Image Clustering with Artifacts Attenuation via Inference-Time Attention Engineering

Improving Image Clustering with Artifacts Attenuation via Inference-Time Attention Engineering

7 Oct 2024arXiv:2410.04801archive 2025-07-28

Kazumoto Nakamura, Yuji Nozawa, Yu-Chieh Lin, Kengo Nakata, Youyang Ng

The goal of this paper is to improve the performance of pretrained Vision Transformer (ViT) models, particularly DINOv2, in image clustering task without requiring re-training or fine-tuning. As model size increases, high-norm artifacts anomaly appears in the patches of multi-head attention. We observe that this anomaly leads to reduced accuracy in zero-shot image clustering. These artifacts are characterized by disproportionately large values in the attention map compared to other patch tokens. To address these artifacts, we propose an approach called Inference-Time Attention Engineering (ITAE), which manipulates attention function during inference. Specifically, we identify the artifacts by investigating one of the Query-Key-Value (QKV) patches in the multi-head attention and attenuate their corresponding attention values inside the pretrained models. ITAE shows improved clustering accuracy on multiple datasets by exhibiting more expressive features in latent space. Our findings highlight the potential of ITAE as a practical solution for reducing artifacts in pretrained ViT models and improving model performance in clustering tasks without the need for re-training or fine-tuning.

Paper PDF

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

Code

No code repository is listed for this paper in the archive or in Syntology's graph.

Code Syntology ran Syntology

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

Tasks

Clustering Image Clustering

Results from the paper archive 2025-07-28

Task	Dataset	Model	Metric	Value	Rank at snapshot	Leaderboard	Report
Image Clustering	CIFAR-10	ITAE	ARI	0.7946	#23 of 40	Archive leaderboard	report
Image Clustering	CIFAR-10	ITAE	Accuracy	0.8449	#23 of 40	Archive leaderboard	report
Image Clustering	CIFAR-10	ITAE	Backbone	ViT-B/14	#23 of 40	Archive leaderboard	report
Image Clustering	CIFAR-10	ITAE	NMI	0.8682	#23 of 40	Archive leaderboard	report
Image Clustering	CIFAR-10	ITAE	Train set	Test	#23 of 40	Archive leaderboard	report
Image Clustering	CIFAR-100	ITAE	ARI	0.5053	#6 of 30	Archive leaderboard	report
Image Clustering	CIFAR-100	ITAE	Accuracy	0.6502	#6 of 30	Archive leaderboard	report
Image Clustering	CIFAR-100	ITAE	Backbone	ViT-B/14	#6 of 30	Archive leaderboard	report
Image Clustering	CIFAR-100	ITAE	NMI	0.771	#6 of 30	Archive leaderboard	report
Image Clustering	CIFAR-100	ITAE	Train Set	Test	#6 of 30	Archive leaderboard	report
Image Clustering	STL-10	ITAE	ARI	0.7594	#14 of 29	Archive leaderboard	report
Image Clustering	STL-10	ITAE	Accuracy	0.8276	#14 of 29	Archive leaderboard	report
Image Clustering	STL-10	ITAE	Backbone	ViT-B/14	#14 of 29	Archive leaderboard	report
Image Clustering	STL-10	ITAE	NMI	0.8818	#14 of 29	Archive leaderboard	report
Image Clustering	STL-10	ITAE	Train Split	Test	#14 of 29	Archive leaderboard	report
Image Clustering	Tiny-ImageNet	ITAE	ARI	0.5227	#2 of 14	Archive leaderboard	report
Image Clustering	Tiny-ImageNet	ITAE	Accuracy	0.6823	#2 of 14	Archive leaderboard	report
Image Clustering	Tiny-ImageNet	ITAE	NMI	0.8178	#2 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

Absolute Position Encodings Adam Attention BPE Dense Connections Dropout Label Smoothing Layer Normalization Linear Layer Multi-Head Attention Position-Wise Feed-Forward Layer Residual Connection Softmax Transformer Vision Transformer

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