AI Paper Atlas

Research map / Detection and segmentation / CNN features

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

Tianqi Li, Wenyu Fang, Xin He, Xue Geng, Xu Cheng, Yun Liu

ECCV 2026 · Detection and segmentation › CNN features · approach: Efficient DETR query design · 1 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
quantize activations to 4 bits (W4A8 to W4A4)CFRN · NC4KS-alpha: .882 -> .443Hurt
quantize weights to 4 bits at 8-bit activations (W8A8 to W4A8)CFRN · NC4KS-alpha: .887 -> .882No clear effect
add DSTG and DCRP together to naive W4A4 (ablation)CFRN · NC4KS-alpha: .443 -> .837Improved
add DSTG only (without DCRP) (ablation)CFRN · NC4KS-alpha: .443 -> .451No clear effect
add DCRP only (without DSTG) (ablation)CFRN · NC4KS-alpha: .443 -> .435Hurt
add DCRP only (without DSTG) (ablation)ESCNet · NC4KS-alpha: .576 -> .771Improved
add DSTG on top of DCRP (ablation)ESCNet · NC4KS-alpha: .771 -> .881Improved
use per-tensor ranges instead of token groups (ablation)CFRN · NC4KS-alpha: .443 -> .407Hurt
use COD-TDQ instead of best PTQ baseline RepQ-ViTCFRN · NC4KS-alpha: .718 -> .837Improved
deploy COD-TDQ W4A4 with INT4 path instead of FP32CFRN · runtimeFPS (img/s): 29.40 -> 44.21Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy