ruby-dnn 0.3.2 → 0.4.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: abf5548b3a0a11d3935f476aa172bab5551598a8c411250ef4816ccca1362d7c
-  data.tar.gz: 1f96a83f5f1ed1e72f10afd1942523f51a80eb32805ee16d10ecc954e7ace669
+  metadata.gz: 9f0d01c2a4696fdf876c6f247a7ea7e41bd57f4d42d2e0065a3703c4faf5a7cc
+  data.tar.gz: 29bcebf1ae90f3b6f72dd391fbba9118162dc49f490b801d1e634f736cc580a4
 SHA512:
-  metadata.gz: 5cb4bfa7d9bea04f287c14bc1e916fa8af45ecf7f3866632d8bd5cf7c3b6fe2a43a64660aade21666b6937ebee2151aa7bd3dbe1f3aa7ea092678d9a9d2c1294
-  data.tar.gz: 2b6887c4b479861956033e3794d9c4415aaf66c572d165c9910addf73f90b3db2d48c51d73178c7f03591ae6bd35080193c8c11479f436332a1eb5e5f59d3cf9
+  metadata.gz: b7b01228b299fbf42ec74f0d4678e6ffd62775a606ecadba3253ce76fd62166699851450d359f13022b01023219b8f710ef75c75af6c1096591df180a49edd62
+  data.tar.gz: 2eb11ae75136aa6be186d81c1fe8c04d585167ce31feccdd26406d58cafd6f333080c871723ad533e9f3ad0fb677c08180ee9925e99040f093f90a08a7f24631

data/API-Reference.ja.md

@@ -2,7 +2,7 @@
 ruby-dnnのAPIリファレンスです。このリファレンスでは、APIを利用するうえで必要となるクラスとメソッドしか記載していません。
 そのため、プログラムの詳細が必要な場合は、ソースコードを参照してください。
 
-最終更新バージョン:0.3.2
+最終更新バージョン:0.4.0
 
 # module DNN
 ruby-dnnの名前空間をなすモジュールです。
@@ -287,13 +287,13 @@ Integer
 
 ## attr_reader :filter_size
 Array   
-フィルターの横と縦の長さ。  
-[Integer width, Integer height]の形式で取得します。
+フィルターの縦と横の長さ。  
+[Integer height, Integer width]の形式で取得します。
 
 ## attr_reader :strides
 Array  
 畳み込みを行う際のストライドの単位。  
-[Integer width, Integer height]の形式で取得します。
+[Integer height, Integer width]の形式で取得します。
 
 ## attr_reader :weight_decay
 Float  
@@ -307,8 +307,8 @@ Float
 * Integer num_filters  
 出力するフィルターの枚数。
 * Integer | Array filter_size  
-フィルターの横と縦の長さ。
-Arrayで指定する場合、[Integer width, Integer height]の形式で指定します。
+フィルターの縦と横の長さ。
+Arrayで指定する場合、[Integer height, Integer width]の形式で指定します。
 * Initializer weight_initializer: nil  
 重みの初期化に使用するイニシャライザーを設定します
 nilを指定すると、RandomNormalイニシャライザーが使用されます。  
@@ -316,7 +316,7 @@ nilを指定すると、RandomNormalイニシャライザーが使用されま
 バイアスの初期化に使用するイニシャライザーを設定します。
 * Array<Integer> strides: 1  
 畳み込みを行う際のストライドの単位を指定します。
-Arrayで指定する場合、[Integer width, Integer height]の形式で指定します。
+Arrayで指定する場合、[Integer height, Integer width]の形式で指定します。
 * bool padding: true  
 イメージに対してゼロパディングを行うか否かを設定します。trueを設定すると、出力されるイメージのサイズが入力されたイメージと同じになるように
 ゼロパディングを行います。
@@ -331,13 +331,13 @@ maxプーリングを行うレイヤーです。
 
 ## attr_reader :pool_size
 Array   
-プーリングを行う横と縦の長さ。
-[Integer width, Integer height]の形式で取得します。
+プーリングを行う縦と横の長さ。
+[Integer height, Integer width]の形式で取得します。
 
 ## attr_reader :strides
 Array  
 畳み込みを行う際のストライドの単位。  
-[Integer width, Integer height]の形式で取得します。
+[Integer height, Integer width]の形式で取得します。
 
 ## 【Instance methods】
 
@@ -345,11 +345,11 @@ Array
 コンストラクタ。
 ### arguments
 * Integer | Array pool_size  
-プーリングを行う横と縦の長さ。
-Arrayで指定する場合、[Integer width, Integer height]の形式で指定します。
+プーリングを行う縦と横の長さ。
+Arrayで指定する場合、[Integer height, Integer width]の形式で指定します。
 * Array<Integer> strides: nil  
 畳み込みを行う際のストライドの単位を指定します。
-Arrayで指定する場合、[Integer width, Integer height]の形式で指定します。
+Arrayで指定する場合、[Integer height, Integer width]の形式で指定します。
 なお、nilが設定された場合は、pool_sizeがstridesの値となります。
 * bool padding: true  
 イメージに対してゼロパディングを行うか否かを設定します。trueを設定すると、出力されるイメージのサイズが入力されたイメージと同じになるように
@@ -363,8 +363,8 @@ Arrayで指定する場合、[Integer width, Integer height]の形式で指定
 
 ## attr_reader :unpool_size
 Array   
-逆プーリングを行う横と縦の長さ。
-[Integer width, Integer height]の形式で取得します。
+逆プーリングを行う縦と横の長さ。
+[Integer height, Integer width]の形式で取得します。
 
 ## 【Instance methods】
 
@@ -372,8 +372,8 @@ Array
 コンストラクタ。
 ### arguments
 * Integer unpool_size  
-逆プーリングを行う横と縦の長さ。
-Arrayで指定する場合、[Integer width, Integer height]の形式で指定します。
+逆プーリングを行う縦と横の長さ。
+Arrayで指定する場合、[Integer height, Integer width]の形式で指定します。
 
 
 # class Flatten

data/LIB-API-Reference.ja.md

@@ -1,6 +1,6 @@
 # LIB-APIリファレンス
 ruby-dnnの付属ライブラリのリファレンスです。  
-最終更新バージョン:0.3.0
+最終更新バージョン:0.4.0
 
 
 # dnn/lib/mnist
@@ -66,7 +66,7 @@ Array
 Array  
 [イメージデータ, ラベルデータ]の形式で取得します。
 * イメージデータ
-  UInt8の[10000, 3, 32, 32]の形式
+  UInt8の[10000, 32, 32, 3]の形式
 * テストデータ
   UInt8の[10000]の形式
 
@@ -78,20 +78,20 @@ Array
 # module ImageIO
 
 ## def self.read(file_name)
-画像をXumo::UInt8形式で読み込みます。
+画像をUInt8形式で読み込みます。
 ### arguments
 * String file_name  
 読み込む画像のファイル名。
 ### return
-Xumo::UInt8  
-[width, height, rgb]のXumo::UInt8配列。
+UInt8  
+[height, width, rgb]のUInt8配列。
 
-## def self.write(file_name, nary, quality: 100)
-Xumo::UInt8形式の画像を書き込みます。
+## def self.write(file_name, img, quality: 100)
+UInt8形式の画像を書き込みます。
 ### arguments
 * String file_name  
 書き込む画像のファイル名。
-* Xumo::UInt8  
-[width, height, rgb]のXumo::UInt8配列。
+* UInt8 img
+[height, width, rgb]のUInt8配列。
 * Integer quality: 100
 画像をJPEGで書き込む場合のクオリティ。

data/bin/console

@@ -1,7 +1,7 @@
 #!/usr/bin/env ruby
 
 require "bundler/setup"
-require "ruby/dnn"
+require "dnn"
 
 # You can add fixtures and/or initialization code here to make experimenting
 # with your gem easier. You can also use a different console, if you like.

data/examples/cifar10_example.rb

@@ -12,8 +12,8 @@ CIFAR10 = DNN::CIFAR10
 x_train, y_train = CIFAR10.load_train
 x_test, y_test = CIFAR10.load_test
 
-x_train = SFloat.cast(x_train).transpose(0, 2, 3, 1)
-x_test = SFloat.cast(x_test).transpose(0, 2, 3, 1)
+x_train = SFloat.cast(x_train)
+x_test = SFloat.cast(x_test)
 
 x_train /= 255
 x_test /= 255
@@ -43,6 +43,16 @@ model << Conv2D.new(32, 5, padding: true)
 model << BatchNormalization.new
 model << ReLU.new
 
+model << MaxPool2D.new(2)
+
+model << Conv2D.new(64, 5, padding: true)
+model << BatchNormalization.new
+model << ReLU.new
+
+model << Conv2D.new(64, 5, padding: true)
+model << BatchNormalization.new
+model << ReLU.new
+
 model << Flatten.new
 
 model << Dense.new(512)

data/lib/dnn/core/layers.rb

@@ -161,69 +161,69 @@ module DNN
     end
     
     
-    #private module
-    module Convert
+    #This module is used for convolution.
+    module Conv2DModule
       private
 
-      def im2col(img, out_w, out_h, fil_w, fil_h, strides)
+      def im2col(img, out_h, out_w, fil_h, fil_w, strides)
         bsize = img.shape[0]
         ch = img.shape[3]
-        col = SFloat.zeros(bsize, ch, fil_w, fil_h, out_w, out_h)
+        col = SFloat.zeros(bsize, ch, fil_h, fil_w, out_h, out_w)
         img = img.transpose(0, 3, 1, 2)
         (0...fil_h).each do |i|
-          i_range = (i...(i + strides[1] * out_h)).step(strides[1]).to_a
+          i_range = (i...(i + strides[0] * out_h)).step(strides[0]).to_a
           (0...fil_w).each do |j|
-            j_range = (j...(j + strides[0] * out_w)).step(strides[0]).to_a
-            col[true, true, j, i, true, true] = img[true, true, j_range, i_range]
+            j_range = (j...(j + strides[1] * out_w)).step(strides[1]).to_a
+            col[true, true, i, j, true, true] = img[true, true, i_range, j_range]
           end
         end
-        col.transpose(0, 4, 5, 2, 3, 1).reshape(bsize * out_w * out_h, fil_w * fil_h * ch)
+        col.transpose(0, 4, 5, 2, 3, 1).reshape(bsize * out_h * out_w, fil_h * fil_w * ch)
       end
 
-      def col2im(col, img_shape, out_w, out_h, fil_w, fil_h, strides)
-        bsize, img_w, img_h, ch = img_shape
-        col = col.reshape(bsize, out_w, out_h, fil_w, fil_h, ch).transpose(0, 5, 3, 4, 1, 2)
-        img = SFloat.zeros(bsize, ch, img_w, img_h)
+      def col2im(col, img_shape, out_h, out_w, fil_h, fil_w, strides)
+        bsize, img_h, img_w, ch = img_shape
+        col = col.reshape(bsize, out_h, out_w, fil_h, fil_w, ch).transpose(0, 5, 3, 4, 1, 2)
+        img = SFloat.zeros(bsize, ch, img_h, img_w)
         (0...fil_h).each do |i|
-          i_range = (i...(i + strides[1] * out_h)).step(strides[1]).to_a
+          i_range = (i...(i + strides[0] * out_h)).step(strides[0]).to_a
           (0...fil_w).each do |j|
-            j_range = (j...(j + strides[0] * out_w)).step(strides[0]).to_a
-            img[true, true, j_range, i_range] += col[true, true, j, i, true, true]
+            j_range = (j...(j + strides[1] * out_w)).step(strides[1]).to_a
+            img[true, true, i_range, j_range] += col[true, true, i, j, true, true]
           end
         end
         img.transpose(0, 2, 3, 1)
       end
 
       def padding(img, pad)
-        bsize, img_w, img_h, ch = img.shape
-        img2 = SFloat.zeros(bsize, img_w + pad[0], img_h + pad[1], ch)
-        i_begin = pad[1] / 2
+        bsize, img_h, img_w, ch = img.shape
+        img2 = SFloat.zeros(bsize, img_h + pad[0], img_w + pad[1], ch)
+        i_begin = pad[0] / 2
         i_end = i_begin + img_h
-        j_begin = pad[0] / 2
+        j_begin = pad[1] / 2
         j_end = j_begin + img_w
-        img2[true, j_begin...j_end, i_begin...i_end, true] = img
+        img2[true, i_begin...i_end, j_begin...j_end, true] = img
         img2
       end
 
       def back_padding(img, pad)
-        i_begin = pad[1] / 2
-        i_end = img.shape[2] - (pad[1] / 2.0).round
-        j_begin = pad[0] / 2
-        j_end = img.shape[1] - (pad[0] / 2.0).round
-        img[true, j_begin...j_end, i_begin...i_end, true]
+        i_begin = pad[0] / 2
+        i_end = img.shape[1] - (pad[0] / 2.0).round
+        j_begin = pad[1] / 2
+        j_end = img.shape[2] - (pad[1] / 2.0).round
+        img[true, i_begin...i_end, j_begin...j_end, true]
       end
 
-      def out_size(prev_w, prev_h, fil_w, fil_h, strides)
-        out_w = (prev_w - fil_w) / strides[0] + 1
-        out_h = (prev_h - fil_h) / strides[1] + 1
-        [out_w, out_h]
+      def out_size(prev_h, prev_w, fil_h, fil_w, strides)
+        out_h = (prev_h - fil_h) / strides[0] + 1
+        out_w = (prev_w - fil_w) / strides[1] + 1
+        [out_h, out_w]
       end
     end
     
     
     class Conv2D < HasParamLayer
       include Initializers
-      include Convert
+      include Conv2DModule
 
       attr_reader :num_filters
       attr_reader :filter_size
@@ -257,12 +257,12 @@ module DNN
 
       def build(model)
         super
-        prev_w, prev_h = prev_layer.shape[0..1]
-        @out_size = out_size(prev_w, prev_h, *@filter_size, @strides)
+        prev_h, prev_w = prev_layer.shape[0..1]
+        @out_size = out_size(prev_h, prev_w, *@filter_size, @strides)
         out_w, out_h = @out_size
         if @padding
-          @pad = [prev_w - out_w, prev_h - out_h]
-          @out_size = [prev_w, prev_h]
+          @pad = [prev_h - out_h, prev_w - out_w]
+          @out_size = [prev_h, prev_w]
         end
       end
 
@@ -317,7 +317,7 @@ module DNN
     
     
     class MaxPool2D < Layer
-      include Convert
+      include Conv2DModule
 
       attr_reader :pool_size
       attr_reader :strides
@@ -341,11 +341,11 @@ module DNN
         super
         prev_w, prev_h = prev_layer.shape[0..1]
         @num_channel = prev_layer.shape[2]
-        @out_size = out_size(prev_w, prev_h, *@pool_size, @strides)
+        @out_size = out_size(prev_h, prev_w, *@pool_size, @strides)
         out_w, out_h = @out_size
         if @padding
-          @pad = [prev_w - out_w, prev_h - out_h]
-          @out_size = [prev_w, prev_h]
+          @pad = [prev_h - out_h, prev_w - out_w]
+          @out_size = [prev_h, prev_w]
         end
       end
 
@@ -383,8 +383,6 @@ module DNN
 
 
     class UnPool2D < Layer
-      include Convert
-
       attr_reader :unpool_size
 
       def initialize(unpool_size)
@@ -398,25 +396,25 @@ module DNN
 
       def build(model)
         super
-        prev_w, prev_h = prev_layer.shape[0..1]
-        unpool_w, unpool_h = @unpool_size
-        out_w = prev_w * unpool_w
+        prev_h, prev_w = prev_layer.shape[0..1]
+        unpool_h, unpool_w = @unpool_size
         out_h = prev_h * unpool_h
-        @out_size = [out_w, out_h]
+        out_w = prev_w * unpool_w
+        @out_size = [out_h, out_w]
         @num_channel = prev_layer.shape[2]
       end
 
       def forward(x)
         @x_shape = x.shape
-        unpool_w, unpool_h = @unpool_size
-        x2 = SFloat.zeros(x.shape[0], x.shape[1], unpool_w, x.shape[2], unpool_h, @num_channel)
+        unpool_h, unpool_w = @unpool_size
+        x2 = SFloat.zeros(x.shape[0], x.shape[1], unpool_h, x.shape[2], unpool_w, @num_channel)
         x2[true, true, 0, true, 0, true] = x
         x2.reshape(x.shape[0], *@out_size, x.shape[3])
       end
 
       def backward(dout)
-        unpool_w, unpool_h = @unpool_size
-        dout = dout.reshape(dout.shape[0], @x_shape[0], unpool_w, @x_shape[1], unpool_h, @num_channel)
+        unpool_h, unpool_w = @unpool_size
+        dout = dout.reshape(dout.shape[0], @x_shape[0], unpool_h, @x_shape[1], unpool_w, @num_channel)
         dout[true, true, 0, true, 0, true].clone
       end
 

data/lib/dnn/ext/dataset_loader/dataset_loader.c

@@ -17,7 +17,7 @@ VALUE mnist_load_images(VALUE self, VALUE rb_bin, VALUE rb_num_images, VALUE rb_
   VALUE rb_na;
   narray_data_t* na_data;
 
-  sprintf(script, "Xumo::UInt8.zeros(%d, %d, %d)", num_images, cols, rows);
+  sprintf(script, "Numo::UInt8.zeros(%d, %d, %d)", num_images, cols, rows);
   rb_na = rb_eval_string(&script[0]);
   na_data = RNARRAY_DATA(rb_na);
 
@@ -35,7 +35,7 @@ VALUE mnist_load_labels(VALUE self, VALUE rb_bin, VALUE rb_num_labels) {
   VALUE rb_na;
   narray_data_t* na_data;
 
-  sprintf(script, "Xumo::UInt8.zeros(%d)", num_labels);
+  sprintf(script, "Numo::UInt8.zeros(%d)", num_labels);
   rb_na = rb_eval_string(&script[0]);
   na_data = RNARRAY_DATA(rb_na);
 
@@ -57,13 +57,13 @@ VALUE cifar10_load(VALUE self, VALUE rb_bin, VALUE rb_num_datas) {
   int k = 0;
   int size = CIFAR10_WIDTH * CIFAR10_HEIGHT * CIFAR10_CHANNEL;
 
-  sprintf(script, "Xumo::UInt8.zeros(%d, %d, %d, %d)", num_datas, CIFAR10_CHANNEL, CIFAR10_WIDTH, CIFAR10_HEIGHT);
+  sprintf(script, "Numo::UInt8.zeros(%d, %d, %d, %d)", num_datas, CIFAR10_CHANNEL, CIFAR10_WIDTH, CIFAR10_HEIGHT);
   rb_na_x = rb_eval_string(&script[0]);
   na_data_x = RNARRAY_DATA(rb_na_x);
   for(i = 0; i < 64; i++) {
     script[i] = 0;
   }
-  sprintf(script, "Xumo::UInt8.zeros(%d)", num_datas);
+  sprintf(script, "Numo::UInt8.zeros(%d)", num_datas);
   rb_na_y = rb_eval_string(&script[0]);
   na_data_y = RNARRAY_DATA(rb_na_y);
 

data/lib/dnn/ext/rb_stb_image/extconf.rb

@@ -0,0 +1,3 @@
+require "mkmf"
+
+create_makefile("rb_stb_image")

data/lib/dnn/ext/rb_stb_image/rb_stb_image.c

@@ -0,0 +1,99 @@
+#include <ruby.h>
+#include "numo/narray.h"
+
+#define STB_IMAGE_IMPLEMENTATION
+#define STB_IMAGE_WRITE_IMPLEMENTATION
+
+#include "stb_image.h"
+#include "stb_image_write.h"
+
+//STBIDEF stbi_uc *stbi_load(char const *filename, int *x, int *y, int *comp, int req_comp);
+VALUE rb_stbi_load(VALUE self, VALUE rb_filename, VALUE rb_req_comp) {
+  char* filename = StringValuePtr(rb_filename);
+  int x, y, n;
+  int req_comp = FIX2INT(rb_req_comp);
+  unsigned char* pixels;
+  narray_data_t* na_data;
+  char script[64];
+  int ch;
+  VALUE rb_x, rb_y, rb_n;
+  VALUE rb_pixels;
+
+  pixels = stbi_load(filename, &x, &y, &n, req_comp);
+  rb_x = INT2FIX(x);
+  rb_y = INT2FIX(y);
+  rb_n = INT2FIX(n);
+  ch = req_comp == 0 ? n : req_comp;
+  sprintf(script, "Numo::UInt8.zeros(%d, %d, %d)", y, x, ch);
+  rb_pixels = rb_eval_string(&script[0]);
+  na_data = RNARRAY_DATA(rb_pixels);
+  memcpy(na_data->ptr, pixels, na_data->base.size);
+  stbi_image_free(pixels);
+  return rb_ary_new3(4, rb_pixels, rb_x, rb_y, rb_n);
+}
+
+//STBIWDEF int stbi_write_png(char const *filename, int w, int h, int comp, const void  *data, int stride_in_bytes);
+VALUE rb_stbi_write_png(VALUE self, VALUE rb_filename, VALUE rb_w, VALUE rb_h, VALUE rb_comp, VALUE rb_pixels, VALUE rb_stride_in_bytes) {
+  char* filename = StringValuePtr(rb_filename);
+  int w = FIX2INT(rb_w);
+  int h = FIX2INT(rb_h);
+  int comp = FIX2INT(rb_comp);
+  unsigned char* pixels;
+  int stride_in_bytes = FIX2INT(rb_stride_in_bytes);
+  narray_data_t* na_data;
+  int result;
+
+  na_data = RNARRAY_DATA(rb_pixels);
+  pixels = (unsigned char*)malloc(na_data->base.size);
+  memcpy(pixels, na_data->ptr, na_data->base.size);
+  result = stbi_write_png(filename, w, h, comp, pixels, stride_in_bytes);
+  stbi_image_free(pixels);
+  return INT2FIX(result);
+}
+
+//STBIWDEF int stbi_write_bmp(char const *filename, int w, int h, int comp, const void  *data);
+VALUE rb_stbi_write_bmp(VALUE self, VALUE rb_filename, VALUE rb_w, VALUE rb_h, VALUE rb_comp, VALUE rb_pixels) {
+  char* filename = StringValuePtr(rb_filename);
+  int w = FIX2INT(rb_w);
+  int h = FIX2INT(rb_h);
+  int comp = FIX2INT(rb_comp);
+  unsigned char* pixels;
+  narray_data_t* na_data;
+  int result;
+
+  na_data = RNARRAY_DATA(rb_pixels);
+  pixels = (unsigned char*)malloc(na_data->base.size);
+  memcpy(pixels, na_data->ptr, na_data->base.size);
+  result = stbi_write_bmp(filename, w, h, comp, pixels);
+  stbi_image_free(pixels);
+  return INT2FIX(result);
+}
+
+//STBIWDEF int stbi_write_jpg(char const *filename, int x, int y, int comp, const void  *data, int quality);
+VALUE rb_stbi_write_jpg(VALUE self, VALUE rb_filename, VALUE rb_w, VALUE rb_h, VALUE rb_comp, VALUE rb_pixels, VALUE rb_quality) {
+  char* filename = StringValuePtr(rb_filename);
+  int w = FIX2INT(rb_w);
+  int h = FIX2INT(rb_h);
+  int comp = FIX2INT(rb_comp);
+  unsigned char* pixels;
+  int quality = FIX2INT(rb_quality);
+  narray_data_t* na_data;
+  int result;
+
+  na_data = RNARRAY_DATA(rb_pixels);
+  pixels = (unsigned char*)malloc(na_data->base.size);
+  memcpy(pixels, na_data->ptr, na_data->base.size);
+  result = stbi_write_jpg(filename, w, h, comp, pixels, quality);
+  stbi_image_free(pixels);
+  return INT2FIX(result);
+}
+
+void Init_rb_stb_image() {
+  VALUE rb_dnn = rb_define_module("DNN");
+  VALUE rb_stb = rb_define_module_under(rb_dnn, "Stb");
+
+  rb_define_module_function(rb_stb, "stbi_load", rb_stbi_load, 2);
+  rb_define_module_function(rb_stb, "stbi_write_png", rb_stbi_write_png, 6);
+  rb_define_module_function(rb_stb, "stbi_write_bmp", rb_stbi_write_bmp, 5);
+  rb_define_module_function(rb_stb, "stbi_write_jpg", rb_stbi_write_jpg, 6);
+}

data/lib/dnn/lib/cifar10.rb

@@ -42,7 +42,9 @@ module DNN
         raise DNN_CIFAR10_LoadError.new(%`file "#{fname}" is not found.`) unless File.exist?(fname)
         bin << File.binread(fname)
       end
-      _cifar10_load(bin, 50000)
+      x_train, y_train = _cifar10_load(bin, 50000)
+      x_train = x_train.transpose(0, 2, 3, 1).clone
+      [x_train, y_train]
     end
 
     def self.load_test
@@ -50,7 +52,9 @@ module DNN
       fname = __dir__ + "/#{CIFAR10_DIR}/test_batch.bin"
       raise DNN_CIFAR10_LoadError.new(%`file "#{fname}" is not found.`) unless File.exist?(fname)
       bin = File.binread(fname)
-      _cifar10_load(bin, 10000)
+      x_test, y_test = _cifar10_load(bin, 10000)
+      x_test = x_test.transpose(0, 2, 3, 1).clone
+      [x_test, y_test]
     end
   end
 end

data/lib/dnn/lib/image_io.rb

@@ -1,26 +1,30 @@
 require "numo/narray"
-require "dnn/ext/image_io/image_io_ext"
+require "dnn/ext/rb_stb_image/rb_stb_image"
 
 module DNN
   module ImageIO
-    private_class_method :_read
-    private_class_method :_write_bmp
-    private_class_method :_write_png
-    private_class_method :_write_jpg
-
     def self.read(file_name)
       raise ImageIO::ReadError.new("#{file_name} is not found.") unless File.exist?(file_name)
-      _read(file_name)
+      img, = Stb.stbi_load(file_name, 3)
+      img
     end
 
-    def self.write(file_name, nary, quality: 100)
+    def self.write(file_name, img, quality: 100)
+      img = img.clone
+      if img.shape.length == 2
+        img = Numo::UInt8[img, img, img].transpose(1, 2, 0).clone
+      elsif img.shape[2] == 1
+        img = img.shape(img.shape[0], img.shape[1])
+        img = Numo::UInt8[img, img, img].transpose(1, 2, 0).clone
+      end
       case file_name
       when /\.png$/
-        _write_png(file_name, nary)
+        stride_in_bytes = img.shape[0] * img.shape[2]
+        Stb.stbi_write_png(file_name, *img.shape, img, stride_in_bytes)
       when /\.bmp$/
-        _write_bmp(file_name, nary)
+        Stb.stbi_write_bmp(file_name, *img.shape, img)
       when /\.jpg$/
-        _write_jpg(file_name, nary, quality)
+        Stb.stbi_write_jpg(file_name, *img.shape, img, quality)
       end
     rescue => ex
       raise ImageIO::WriteError.new(ex.message)

data/lib/dnn/version.rb

@@ -1,3 +1,3 @@
 module DNN
-  VERSION = "0.3.2"
+  VERSION = "0.4.0"
 end

data/ruby-dnn.gemspec

@@ -13,7 +13,7 @@ Gem::Specification.new do |spec|
   spec.description   = %q{ruby-dnn is a ruby deep learning library.}
   spec.homepage      = "https://github.com/unagiootoro/ruby-dnn.git"
   spec.license       = "MIT"
-  spec.extensions    = ["lib/dnn/ext/dataset_loader/extconf.rb", "lib/dnn/ext/image_io/extconf.rb"]
+  spec.extensions    = ["lib/dnn/ext/dataset_loader/extconf.rb", "lib/dnn/ext/rb_stb_image/extconf.rb"]
 
   spec.add_dependency "numo-narray"
   spec.add_dependency "archive-tar-minitar"

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: ruby-dnn
 version: !ruby/object:Gem::Version
-  version: 0.3.2
+  version: 0.4.0
 platform: ruby
 authors:
 - unagiootoro
 autorequire: 
 bindir: exe
 cert_chain: []
-date: 2018-07-20 00:00:00.000000000 Z
+date: 2018-07-21 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: numo-narray
@@ -86,7 +86,7 @@ email:
 executables: []
 extensions:
 - lib/dnn/ext/dataset_loader/extconf.rb
-- lib/dnn/ext/image_io/extconf.rb
+- lib/dnn/ext/rb_stb_image/extconf.rb
 extra_rdoc_files: []
 files:
 - ".gitignore"
@@ -143,39 +143,39 @@ files:
 - lib/dnn/ext/dataset_loader/numo/types/uint8.h
 - lib/dnn/ext/dataset_loader/numo/types/uint_macro.h
 - lib/dnn/ext/dataset_loader/numo/types/xint_macro.h
-- lib/dnn/ext/image_io/extconf.rb
-- lib/dnn/ext/image_io/image_io_ext.c
-- lib/dnn/ext/image_io/numo/compat.h
-- lib/dnn/ext/image_io/numo/extconf.h
-- lib/dnn/ext/image_io/numo/intern.h
-- lib/dnn/ext/image_io/numo/narray.h
-- lib/dnn/ext/image_io/numo/ndloop.h
-- lib/dnn/ext/image_io/numo/template.h
-- lib/dnn/ext/image_io/numo/types/bit.h
-- lib/dnn/ext/image_io/numo/types/complex.h
-- lib/dnn/ext/image_io/numo/types/complex_macro.h
-- lib/dnn/ext/image_io/numo/types/dcomplex.h
-- lib/dnn/ext/image_io/numo/types/dfloat.h
-- lib/dnn/ext/image_io/numo/types/float_def.h
-- lib/dnn/ext/image_io/numo/types/float_macro.h
-- lib/dnn/ext/image_io/numo/types/int16.h
-- lib/dnn/ext/image_io/numo/types/int32.h
-- lib/dnn/ext/image_io/numo/types/int64.h
-- lib/dnn/ext/image_io/numo/types/int8.h
-- lib/dnn/ext/image_io/numo/types/int_macro.h
-- lib/dnn/ext/image_io/numo/types/real_accum.h
-- lib/dnn/ext/image_io/numo/types/robj_macro.h
-- lib/dnn/ext/image_io/numo/types/robject.h
-- lib/dnn/ext/image_io/numo/types/scomplex.h
-- lib/dnn/ext/image_io/numo/types/sfloat.h
-- lib/dnn/ext/image_io/numo/types/uint16.h
-- lib/dnn/ext/image_io/numo/types/uint32.h
-- lib/dnn/ext/image_io/numo/types/uint64.h
-- lib/dnn/ext/image_io/numo/types/uint8.h
-- lib/dnn/ext/image_io/numo/types/uint_macro.h
-- lib/dnn/ext/image_io/numo/types/xint_macro.h
-- lib/dnn/ext/image_io/stb_image.h
-- lib/dnn/ext/image_io/stb_image_write.h
+- lib/dnn/ext/rb_stb_image/extconf.rb
+- lib/dnn/ext/rb_stb_image/numo/compat.h
+- lib/dnn/ext/rb_stb_image/numo/extconf.h
+- lib/dnn/ext/rb_stb_image/numo/intern.h
+- lib/dnn/ext/rb_stb_image/numo/narray.h
+- lib/dnn/ext/rb_stb_image/numo/ndloop.h
+- lib/dnn/ext/rb_stb_image/numo/template.h
+- lib/dnn/ext/rb_stb_image/numo/types/bit.h
+- lib/dnn/ext/rb_stb_image/numo/types/complex.h
+- lib/dnn/ext/rb_stb_image/numo/types/complex_macro.h
+- lib/dnn/ext/rb_stb_image/numo/types/dcomplex.h
+- lib/dnn/ext/rb_stb_image/numo/types/dfloat.h
+- lib/dnn/ext/rb_stb_image/numo/types/float_def.h
+- lib/dnn/ext/rb_stb_image/numo/types/float_macro.h
+- lib/dnn/ext/rb_stb_image/numo/types/int16.h
+- lib/dnn/ext/rb_stb_image/numo/types/int32.h
+- lib/dnn/ext/rb_stb_image/numo/types/int64.h
+- lib/dnn/ext/rb_stb_image/numo/types/int8.h
+- lib/dnn/ext/rb_stb_image/numo/types/int_macro.h
+- lib/dnn/ext/rb_stb_image/numo/types/real_accum.h
+- lib/dnn/ext/rb_stb_image/numo/types/robj_macro.h
+- lib/dnn/ext/rb_stb_image/numo/types/robject.h
+- lib/dnn/ext/rb_stb_image/numo/types/scomplex.h
+- lib/dnn/ext/rb_stb_image/numo/types/sfloat.h
+- lib/dnn/ext/rb_stb_image/numo/types/uint16.h
+- lib/dnn/ext/rb_stb_image/numo/types/uint32.h
+- lib/dnn/ext/rb_stb_image/numo/types/uint64.h
+- lib/dnn/ext/rb_stb_image/numo/types/uint8.h
+- lib/dnn/ext/rb_stb_image/numo/types/uint_macro.h
+- lib/dnn/ext/rb_stb_image/numo/types/xint_macro.h
+- lib/dnn/ext/rb_stb_image/rb_stb_image.c
+- lib/dnn/ext/rb_stb_image/stb_image.h
+- lib/dnn/ext/rb_stb_image/stb_image_write.h
 - lib/dnn/lib/cifar10.rb
 - lib/dnn/lib/image_io.rb
 - lib/dnn/lib/mnist.rb

data/lib/dnn/ext/image_io/extconf.rb

@@ -1,3 +0,0 @@
-require "mkmf"
-
-create_makefile("image_io_ext")

data/lib/dnn/ext/image_io/image_io_ext.c

@@ -1,89 +0,0 @@
-#include <ruby.h>
-#include "numo/narray.h"
-
-#define STB_IMAGE_IMPLEMENTATION
-#define STB_IMAGE_WRITE_IMPLEMENTATION
-
-#include "stb_image.h"
-#include "stb_image_write.h"
-
-VALUE image_io_read(VALUE self, VALUE rb_file_name) {
-  char* file_name = StringValuePtr(rb_file_name);
-  int width;
-  int height;
-  int bpp;
-  unsigned char* pixels;
-  char script[64];
-  VALUE rb_na;
-  narray_data_t* na_data;
-  pixels = stbi_load(file_name, &width, &height, &bpp, 3);
-  sprintf(script, "Xumo::UInt8.zeros(%d, %d, 3)", width, height);
-  rb_na = rb_eval_string((char*)script);
-  na_data = RNARRAY_DATA(rb_na);
-  memcpy(na_data->ptr, pixels, na_data->base.size);
-  stbi_image_free(pixels);
-  return rb_na;
-}
-
-VALUE image_io_write_png(VALUE self, VALUE rb_file_name, VALUE rb_na) {
-  char* file_name = StringValuePtr(rb_file_name);
-  int width;
-  int height;
-  int bpp = 3;
-  unsigned char* pixels;
-  narray_data_t* na_data;
-  na_data = RNARRAY_DATA(rb_na);
-  pixels = (unsigned char*)malloc(na_data->base.size);
-  memcpy(pixels, na_data->ptr, na_data->base.size);
-  width = na_data->base.shape[0];
-  height = na_data->base.shape[1];
-  stbi_write_png(file_name, width, height, bpp, pixels, width * bpp);
-  stbi_image_free(pixels);
-  return Qnil;
-}
-
-VALUE image_io_write_bmp(VALUE self, VALUE rb_file_name, VALUE rb_na) {
-  char* file_name = StringValuePtr(rb_file_name);
-  int width;
-  int height;
-  int bpp = 3;
-  unsigned char* pixels;
-  narray_data_t* na_data;
-  na_data = RNARRAY_DATA(rb_na);
-  pixels = (unsigned char*)malloc(na_data->base.size);
-  memcpy(pixels, na_data->ptr, na_data->base.size);
-  width = na_data->base.shape[0];
-  height = na_data->base.shape[1];
-  stbi_write_bmp(file_name, width, height, bpp, pixels);
-  stbi_image_free(pixels);
-  return Qnil;
-}
-
-VALUE image_io_write_jpg(VALUE self, VALUE rb_file_name, VALUE rb_na, VALUE rb_quality) {
-  char* file_name = StringValuePtr(rb_file_name);
-  int width;
-  int height;
-  int bpp = 3;
-  int quality = FIX2INT(rb_quality);
-  unsigned char* pixels;
-  narray_data_t* na_data;
-  na_data = RNARRAY_DATA(rb_na);
-  pixels = (unsigned char*)malloc(na_data->base.size);
-  memcpy(pixels, na_data->ptr, na_data->base.size);
-  width = na_data->base.shape[0];
-  height = na_data->base.shape[1];
-  stbi_write_jpg(file_name, width, height, bpp, pixels, quality);
-  stbi_image_free(pixels);
-  return Qnil;
-}
-
-void Init_image_io_ext() {
-  VALUE rb_dnn;
-  VALUE rb_image_io;
-  rb_dnn = rb_define_module("DNN");
-  rb_image_io = rb_define_module_under(rb_dnn, "ImageIO");
-  rb_define_singleton_method(rb_image_io, "_read", image_io_read, 1);
-  rb_define_singleton_method(rb_image_io, "_write_png", image_io_write_bmp, 2);
-  rb_define_singleton_method(rb_image_io, "_write_bmp", image_io_write_png, 2);
-  rb_define_singleton_method(rb_image_io, "_write_jpg", image_io_write_jpg, 3);
-}