存档
Berkeley计算机视觉公开课

Computer vision seeks to develop algorithms that replicate one of the most amazing capabilities of the human brain – inferring properties of the external world purely by means of the light reflected from various objects to the eyes. We can determine how far away these objects are, how they are oriented with respect to us, and in relationship to various other objects. We reliably guess their colors and textures, and we can recognize them – this is a chair, this is my dog Fido, this is a picture of Bill Clinton smiling. We can segment out regions of space corresponding to particular objects and track them over time, such as a basketball player weaving through the court.
kinect video 6/100: DarwinBot
Starry Night Visualization
Goggles官方算法介绍
google在ICML2011上做的关于goggles的介绍,非常值得一看。
演讲人是google的Hartmut Neven,是负责visual search的老大?
欢迎大家努力发掘相关信息。
视频最后提到了用到了量子计算。(表示完全不懂,求扫盲)。
Kmeans based indexing and Asymmetric Distance Computation for ANN search (Binary Local Feature): part1
受Herve Jegou的Hamming Embedding and Weak Geometric consistency for large-scale image search以及Product quantization for nearest neighbor search的启发,将Kmeans clustering、inverted files、Asymmetric Distance Computation应用到二进制形式的局部特征的最近邻检索。
主要思路:
用Kmeans做特征的粗索引。
根据统计数据对feature进行压缩。
检索时使用非对称的方式计算索引特征与查询特征之间的距离。
算法:
训练:
- 使用Kmeans对欲索引的特征进行聚类,得到K个中心。对二进制形式的feature做聚类时,类别中心更新方式为:对于每一个bit,统计所有落在该类别的特征的对应bit上的1,0频率,并取高者。
- 对于每个cluster,统计所有落在该类别的特征的每个bit位的1,0频率,取1或者0频率靠近50%的前M个bits。(越靠近50%,熵越大)
经过训练,我们得到两组数据:
- K个特征类别中心。
- 对于每个类别中心,都有一组“M个bit位置标示符”。这些标示符构成一个对原始feature进行压缩的依据。(本文以后将其称为投影向量)
CV Dazzle

之前介绍过一个反人脸检测的东西,作者Adam Harvey又将其发扬光大了,整出一个CV Dazzle。最早的版本没有考虑到审美上的因素,纯粹只是为了干掉人脸检测器(OpenCV based),这次Adam Harvey试图弄得。。好看点?
最新评论