[论文] FLEX-LBP: Learning Local Binary Patterns with Exact Binary Inference f...
研究领域: CV 作者: Mohamed Alimoussa, Amal El Kaid 发布时间: 2025-11-13 arXiv: 2511.10634
论文概要
研究领域: CV 作者: Mohamed Alimoussa, Amal El Kaid 发布时间: 2025-11-13 arXiv: 2511.10634
中文摘要
局部二值模式(LBP)仍被广泛使用,但其半径、采样角、颜色变换和阈值均按数据集手工固定。FLEX-LBP 通过反向传播学习这些参数,同时每次前向传播计算精确二值码与硬直方图计数。部署的描述子是可解释的硬二值 LBP,仅 90 个可学习参数且推理成本不变。在五个种子上平均将最近邻精度提升 3.55 个点,超过冻结的 ImageNet ResNet-50 特征,在 Kather 结直肠癌组织学基准上达到 93.8%。
原文摘要
Local Binary Patterns remain widely used but their parameters are hand-fixed per dataset. FLEX-LBP learns radius, sampling angles, color transform and thresholds by backpropagation while every forward pass computes exact binary codes and hard histogram counts. The deployed descriptor is an interpretable hard-binary LBP with 90 learnable parameters and unchanged inference cost. It improves nearest-neighbor accuracy by 3.55 points on average over five seeds, exceeds frozen ImageNet ResNet-50 features, and reaches 93.8% on Kather colorectal cancer histology benchmark.
*自动采集于 2026-11-15*
#论文 #arXiv #CV #小凯