CVG Talk - Saliency based detection

最新推荐文章于 2021-08-19 17:08:51 发布

原创最新推荐文章于 2021-08-19 17:08:51 发布 · 501 阅读

0 ·

CC 4.0 BY-SA版权

本文介绍了一种名为BooleanMapSaliency (BMS) 的视觉显著性地图生成方法，该方法首先将RGB图像转换为LAB颜色空间，并通过不同灰度阈值为每个颜色通道生成布尔地图。接着，通过激活映射、归一化和高斯模糊等步骤生成最终的显著性地图。此外，还讨论了用于评估显著性地图质量的各种指标，如平均绝对误差、ROC曲线和精度-召回曲线。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

                    
                    Method

  Albrecht, 2011
BMS
Liu
Makantasis

Boolean Map Saliency (BMS)

  RGB –> LAB
Generate Boolean Maps (thresholded by different grey level, for each colour channel), for delta = 8, in total 3*
Activation Maps
Normalisation
Normalised Activation Maps
combine all together, then  Gaussian blur

Metrics

  Mean Absolute Error
ROC Curve, i.e. TRP, FPR
Precision-Recall Curve

Saliency-bootstrapped Tracking

  Visual saliency bootstrapping
Temporally Stable Feature Cluster Tracking 

  model free
does not require learning
limited long term capability
detection of wakes