The passage discusses the importance of summarization in an age of information overload. It notes that with the massive amounts of data available online, being able to quickly understand the key points of lengthy documents, articles, or reports is crucial. The ability to produce clear, concise summaries helps people filter through large amounts of information and identify what is most important or relevant to them.
- The document summarizes a presentation on deep learning and neural networks given by Junya Kaneko of MPS Yokohama.
- It introduces key concepts of deep learning, neural networks, and TensorFlow for deep learning.
- It provides examples of applications of deep learning like in Android OS, Google Photos, Facebook's DeepFace, and self-driving cars.
- The presentation demonstrates how to set up TensorFlow on a virtual machine and build simple neural networks.
15. 離散化 Gaussian Convolution の計算量
Mi: m 行 n 列の行列, G: s 行 t 列の行列
Mo の1画素を求めるために必要な計算量?
- s * t
Mo の全画素を求めるために必要な計算量?
- m * n * s * t
例: 縦横256ピクセルの画像と縦横12ピクセルのフィルタ?
? 256 * 256 * 12 * 12 = 9437184 ケロ???
MPS 第38回 (2015/11/14) 定例ミーティング資料 (c) Junya Kaneko
21. 離散化 Gaussian Convolution の計算量再考
Mi: m 行 n 列の行列, G: s 行 t 列の行列
Mo の前画素を求めるために必要な計算量
- m * n* t + m * n * s = m * n * (s + t)
例: 縦横256ピクセルの画像と縦横12ピクセルのフィルタ?
? 256 * 256 * (12 + 12) = 1572864
? 前回の結果 9437184 と比較すると約17%の計算量
ケロ !!!
MPS 第38回 (2015/11/14) 定例ミーティング資料 (c) Junya Kaneko
44. 参考文献
1. David G. Lowe. 2004. Distinctive Image Features from Scale-Invariant
Keypoints. Int. J. Comput. Vision 60, 2 (November 2004), 91-110.
2. M. Alex O. Vasilescu: http://alumni.media.mit.edu/~maov/classes/
vision09/lect/09_Image_Filtering_Edge_Detection_09.pdf
MPS 第38回 (2015/11/14) 定例ミーティング資料 (c) Junya Kaneko