AI Paper Atlas

Research map / Detection and segmentation / CNN features

Just Noticeable Difference Modeling for Deep Visual Features

Rui Zhao, Wenrui Li, Lin Zhu, Yajing Zheng, Weisi Lin

ICML 2026 · Detection and segmentation › CNN features · approach: Convolution-transformer backbones · 0 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
use FeatJND-shaped distortion instead of Gaussian noise at matched NRMSEfeature-space distortion · ImageNet-1K, COCO (all backbones)task performanceImproved
use FeatJND-guided token-wise step sizes instead of random permutation of step sizestoken-wise feature quantization · ImageNet / COCO, transformer backbonesaccuracy / mAPImproved
use FeatJND-guided quantization instead of global uniform steptoken-wise feature quantization · ImageNet / COCO, transformer backbonesaccuracy / mAPImproved
increase loss weight lambda_t from 10 to 50 (ablation)FeatJND · Swin-S ImageNet-1K, NRMSE 40Top-1 accuracy (%): 53.7 -> 82.5Improved
increase lambda_t from 10 to 200 (ablation)FeatJND · Swin-S + Mask R-CNN detection, NRMSE 5mAP (%): 38.3 -> 47.5Improved
increase lambda_t beyond 50 (50 to 70) (ablation)FeatJND · Swin-S ImageNet-1KTop-1 accuracy (%): 82.5 -> 82.5No clear effect
raise distillation temperature T from 1 to 4 (ablation)FeatJND · Swin-S ImageNet-1K, NRMSE 40Top-1 accuracy (%): 78.7 -> 82.5Improved
raise temperature T from 4 to 8 (ablation)FeatJND · Swin-S ImageNet-1KTop-1 accuracy (%): 82.5 -> 82.4No clear effect
raise temperature T from 1 to 4 (ablation)FeatJND · Swin-S + Mask R-CNN detection, NRMSE 5mAP (%): 45.1 -> 47.5Improved
vary number of residual blocks in estimator (ablation)FeatJND · Swin-S ImageNet-1KTop-1 accuracy (%)No clear effect
use larger 5x5 kernels instead of 1x1 in estimator (ablation)FeatJND · Swin-S ImageNet-1K, NRMSE 40Top-1 accuracy (%): 82.5 -> 81.8Hurt
scale FeatJND strength alpha above 1 (ablation)FeatJND · ImageNet / COCOperformance dropHurt

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