Peter M. Groen / openbr

Commit a039342e8a6718ccfbb4d8574e3122d7d3c8b000

Authored by Josh Klontz
1 parent 39c93c18

likely normalize-l2 now works

Inline Side-by-side
Showing 2 changed files with 21 additions and 17 deletions
openbr/plugins/core/likely.cpp
  View file @a039342
... ... @@ -11,7 +11,8 @@ namespace br
11 11 * \ingroup transforms
12 12 * \brief Generic interface to Likely JIT compiler
13 13 *
14   - * www.liblikely.org
  14 + * - [Homepage](www.liblikely.org)
  15 + * - [API Documentation](https://s3.amazonaws.com/liblikely/doxygen/index.html)
15 16 * \author Josh Klontz \cite jklontz
16 17 */
17 18 class LikelyTransform : public Transform
... ...
share/openbr/likely/face_recognition.tex
  View file @a039342
... ... @@ -21,6 +21,7 @@ As Likely is a literate programming language, this document is both the source c
21 21 \end{abstract}
22 22  
23 23 \section{Compilation}
  24 +\label{sec:compilation}
24 25 This document is a valid \LaTeX\ file that can be rendered for reading by humans.
25 26  
26 27 \begin{verbatim}
... ... @@ -34,9 +35,10 @@ $ likely -p 'data := "path/to/training/data.lm".read-matrix' \
34 35 face_recognition.tex face_recognition.o
35 36 \end{verbatim}
36 37  
37   -Note that the outputted \emph{face\_recognition.o} native object file exposes its functionality through a C-compatible signature.
  38 +Note that the outputted \texttt{face\_recognition.o} native object file exposes its functionality through a C-compatible signature.
38 39  
39 40 \begin{verbatim}
  41 +#include <likely.h>
40 42 likely_mat face_recognition(likely_const_mat src);
41 43 \end{verbatim}
42 44  
... ... @@ -50,10 +52,12 @@ face-recognition :=
50 52 \end{likely}
51 53  
52 54 Throughout the remainder of this document we will append additional steps to this function, building up the entire algorithm one transformation at a time.
53   -It is the responsibility of each section in this document to update the \emph{face-recognition} function and the training data.
54   -In this way the sections maintain independence, and the algorithm can be changed by inserting, removing, or editing sections of this document.
  55 +It is the responsibility of each section in this document to update the \texttt{face-recognition} function and the training \texttt{data}.
  56 +It is in this way that the sections maintain independence, and the algorithm can be changed by inserting, removing, or editing sections of this document.
  57 +Remember that the \emph{Likely Language Reference}\footnote{https://s3.amazonaws.com/liblikely/latex/standard.pdf} is a good resource to consult if you are uncertain about the meaning of some code.
55 58  
56   -Note that the remainder of this document assumes that the global variable \emph{data} is defined, and contains the appropriate training samples.
  59 +Note that the remainder of this document assumes that the global variable
  60 +\texttt{data} is defined, and contains the appropriate training samples.
57 61  
58 62 \begin{likely}
59 63 "Number of training samples:"
... ... @@ -73,19 +77,18 @@ Such a normalization has been observed to improve accuracy when comparing featur
73 77 normalize-l2 :=
74 78 src :->
75 79 {
76   - norm := (new f32T src.channels 1 src.rows src.frames null)
77   - (norm src) :=>
  80 + dst := src.imitate
  81 + len := src.columns
  82 + ((src.channels 1 src.rows src.frames)
  83 + dst src len) :=>
78 84 {
79   - add-squared-element :=
80   - x :->
81   - norm :<- (+ norm src.f32.sq)
82   - (iter add-squared-element src.columns)
83   - norm :<- (/ 1 (sqrt norm))
  85 + norm := 0.(cast src).$
  86 + (iter (-> x (<- norm (+ norm src.sq)))
  87 + len)
  88 + norm := (/ 1 (sqrt norm))
  89 + (iter (-> x (<- dst (* src norm)))
  90 + len)
84 91 }
85   -
86   - dst := src.imitate
87   - (dst src norm) :=>
88   - dst :<- (* src norm)
89 92 }
90 93 \end{likely}
91 94  
... ... @@ -149,7 +152,7 @@ data.columns
149 152 \end{likely}
150 153  
151 154 \section{Entry Point}
152   -The entry point to the outside world is a single C function \texttt{face\_recognition} that takes as input a \texttt{likely\_mat} (the input image) and outputs a new \texttt{likely\_mat} (the feature vector).
  155 +Recall from section \ref{sec:compilation} that the entry point to the outside world is a single C function \texttt{face\_recognition} that takes as input a \texttt{likely\_mat} (the input image) and outputs a new \texttt{likely\_mat} (the feature vector). Now that we have finished defining \texttt{face-recognition} we can finally export this symbol.
153 156  
154 157 \begin{likely}
155 158 (extern u8X "face_recognition" f32X face-recognition)
... ...