When W4A4 Breaks Camouflaged Object Detection: Token-Group Dual-Constraint Activation Quantization

2026-04-18

AI Summary

This work addresses the severe degradation of object boundary details in 4-bit activation quantization for camouflaged object detection (COD), caused by background-dominated tokens skewing the shared activation range and pushing critical features into the zero quantization bin. It is the first to identify this token-local bottleneck inherent to low-bit quantization in COD tasks. To mitigate this, the authors propose COD-TDQ, which employs Direct-Sum Token grouping to assign localized scales, thereby suppressing cross-token range dominance, and introduces a Dual-Constraint Range Projection mechanism that jointly optimizes the clipping range to balance step size–discretization ratio and zero-bin quality. Without requiring retraining, COD-TDQ achieves consistent gains across four COD benchmarks and two backbone architectures, improving the Sα metric by over 0.12 compared to state-of-the-art quantization methods and substantially enhancing 4-bit inference performance.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Object Detection & Categorization
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

Camouflaged object detection (COD) segments objects that intentionally blend with the background, so predictions depend on subtle texture and boundary cues. COD is often needed under tight on-device memory and latency budgets, making low-bit inference highly desirable. However, COD is unusually hard to quantify aggressively. We study post-training W4A4 quantization of Transformer-based COD and find a task-specific cliff: heavy-tailed background tokens dominate a shared activation range, inflating the step size and pushing weak-but-structured boundary cues into the zero bin. This exposes a token-local bottleneck -- remove cross-token range domination and bound the zero-bin mass under 4-bit activations. To address this, we introduce COD-TDQ, a COD-aware Token-group Dual-constraint activation Quantization method. COD-TDQ addresses this token-local bottleneck with two coupled steps: Direct-Sum Token-Group (DSTG) assigns token-group scales to suppress cross-token range domination, and Dual-Constraint Range Projection (DCRP) projects each token-group clip range to keep the step-to-dispersion ratio and the zero-bin mass bounded. Across four COD benchmarks and two baseline models (CFRN and ESCNet), COD-TDQ consistently achieves an Sαscore more than 0.12 higher than that of the state-of-the-art quantization method without retraining. The code will be released.

Problem

• Camouflaged Object Detection
• Low-bit Quantization
• Activation Quantization
• Token-level Bottleneck
• W4A4

Innovation

• Camouflaged Object Detection
• W4A4 Quantization
• Token-Group Quantization
• Activation Quantization
• Zero-Bin Suppression