Efficient Input-Level Backdoor Defense on Text-to-Image Synthesis via Neuron Activation Variation
Authors:
- 19 October 2025
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 15182-15193
Abstract
In recent years, text-to-image (T2I) diffusion models have gained significant attention for their ability to generate highquality images reflecting text prompts. However, their growing popularity has also led to the emergence of backdoor threats, posing substantial risks. Currently, effective defense strategies against such threats are lacking due to the diversity of backdoor targets in T2I synthesis. In this paper, we propose NaviT2I, an efficient input-level backdoor defense framework against diverse T2I backdoors. Our approach is based on the new observation that trigger tokens tend to induce significant neuron activation variation in the early stage of the diffusion generation process, a phenomenon we term Early-step Activation Variation. Leveraging this insight, NaviT2I navigates T2I models to prevent malicious inputs by analyzing Neuron activation variations caused by input tokens. Extensive experiments show that NaviT2I significantly outperforms the baselines in both effectiveness and efficiency across diverse datasets, various T2I backdoors, and different model architectures including UNet and DiT. Furthermore, we show that our method remains effective under potential adaptive attacks.11Our code will be released at: https://github.com/zhaisf/NaviT2I.Keywords
All Related Versions
Funding Information
- National Key Research and Development Program of China (2022YFB2703301)
- National Science Foundation (CNS-2424127)
This publication has 49 references indexed in Scilit:
- T2IShield: Defending Against Backdoors on Text-to-Image Diffusion Models2024
- BadDet: Backdoor Attacks on Object Detection2023
- Data-Free Backdoor Removal Based on Channel Lipschitzness2022
- Kallima: A Clean-Label Framework for Textual Backdoor Attacks2022
- The Devil Is in the GAN: Backdoor Attacks and Defenses in Deep Generative Models2022
- Mitigating backdoor attacks in LSTM-based text classification systems by Backdoor Keyword IdentificationNeurocomputing, 2021
- Fine-Pruning: Defending Against Backdooring Attacks on Deep Neural Networks2018
- U-Net: Convolutional Networks for Biomedical Image Segmentation2015
- Microsoft COCO: Common Objects in Context2014
- An introduction to ROC analysisPattern Recognition Letters, 2006