0656706

results/index.md

@@ -1,47 +1,85 @@
-# Your Name <span style="color:red">(id)</span>
+# 楊皓崴 <span style="color:red">(0656706)</span>
 
-#Project 5: Deep Classification
+#Project 1: Deep Classification
 
 ## Overview
 The project is related to 
-> quote
+> [ImageNet Classification with Deep Convolutional Neural Networks](https://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks)
 
 
 ## Implementation
-1. One
-	* item
-	* item
-2. Two
+1. 前處理
+圖檔過大容易耗盡deepQ資源，故將長寬縮小24倍(45*80)  
+整理成一般分類問題資料儲存之  
 
-```
-Code highlights
-```
+```python
+    file = os.path.join(path4,"Image"+str(fileN)+".png")
+    pngfile = Image.open(file)
+    pngfile.thumbnail(size, Image.ANTIALIAS)
+    temp = np.asarray( pngfile, dtype="uint8" )
+``` 
+
+2. 模型結構
+    * conv layers (3*3 filter) *3
+    * fully connect *2
+
+```python
+	def alex_net(_X, _weights, _biases, _dropout):
+    	_X = tf.reshape(_X, shape=[-1, 45, 80, 3])
+
+    	conv1 = conv2d('conv1', _X, _weights['wc1'], _biases['bc1'])
+    	pool1 = max_pool('pool1', conv1, k=2)
+    	norm1 = norm('norm1', pool1, lsize=4)
+    	norm1 = tf.nn.dropout(norm1, _dropout)
+
+	    conv2 = conv2d('conv2', norm1, _weights['wc2'], _biases['bc2'])
+    	pool2 = max_pool('pool2', conv2, k=2)
+    	norm2 = norm('norm2', pool2, lsize=4)
+    	norm2 = tf.nn.dropout(norm2, _dropout)
+
+    	conv3 = conv2d('conv3', norm2, _weights['wc3'], _biases['bc3'])
+    	pool3 = max_pool('pool3', conv3, k=2)
+    	norm3 = norm('norm3', pool3, lsize=4)
+    	norm3 = tf.nn.dropout(norm3, _dropout)
+
+    	dense1 = tf.reshape(norm3, [-1, _weights['wd1'].get_shape().as_list()[0]])
+    	dense1 = tf.nn.relu(tf.matmul(dense1, _weights['wd1']) + _biases['bd1'], name='fc1')
+
+    	dense2 = tf.nn.relu(tf.matmul(dense1, _weights['wd2']) + _biases['bd2'], name='fc2')
+
+    	out = tf.matmul(dense2, _weights['out']) + _biases['out']
+    	return out
+``` 
 
 ## Installation
-* Other required packages.
-* How to compile from source?
+* Required packages.
+    * tensorflow
+    * numpy
+* How to compile from source?  
+    * python preprocessing.py
+    * python process01.py / python process02.py
 
 ### Results
 
 <table border=1>
 <tr>
 <td>
-<img src="placeholder.jpg" width="24%"/>
-<img src="placeholder.jpg"  width="24%"/>
-<img src="placeholder.jpg" width="24%"/>
-<img src="placeholder.jpg" width="24%"/>
+process01.py
+</td>
+<td>
+process02.py
 </td>
 </tr>
 
 <tr>
 <td>
-<img src="placeholder.jpg" width="24%"/>
-<img src="placeholder.jpg"  width="24%"/>
-<img src="placeholder.jpg" width="24%"/>
-<img src="placeholder.jpg" width="24%"/>
+<img src="https://github.com/w95wayne10/homework1/blob/master/results/output01.PNG" width="80%"/>
+</td>
+<td>
+<img src="https://github.com/w95wayne10/homework1/blob/master/results/output02.PNG"  width="80%"/>
 </td>
 </tr>
 
 </table>
 
-
+嘗試增加conv層數，但沒有效果，猜測圖片縮小過多所致，但操作時間過久來不及嘗試其他尺寸..

results/preprocessing.py

@@ -0,0 +1,74 @@
+import numpy as np
+from PIL import Image
+import os
+
+path0 = "frames"
+testData = None
+trainData = None
+testLabel = None
+trainLabel = None
+size = 1920/24, 1080/24
+lr = None
+for train_test in os.listdir(path0):
+    path1 = os.path.join(path0, train_test)
+    for site in os.listdir(path1):
+        path2 = os.path.join(path1, site)
+        for num in os.listdir(path2):
+            path3 = os.path.join(path2, num)
+            for LR in os.listdir(path3):
+                if LR == "head":
+                    continue
+                elif LR == "Lhand":
+                    lr = "left"
+                else:
+                    lr = "right"
+                path4 = os.path.join(path3, LR)
+                print(path4)
+
+                for fileN in range(1,len(os.listdir(path4))+1):
+                    file = os.path.join(path4,"Image"+str(fileN)+".png")
+                    pngfile = Image.open(file)
+                    pngfile.thumbnail(size, Image.ANTIALIAS)
+                    temp = np.asarray( pngfile, dtype="uint8" )
+                    if train_test == "train":
+                        if trainData is None:
+                            trainData = temp.reshape([1,-1])
+                        else:
+                            trainData = np.concatenate([trainData,temp.reshape([1,-1])])
+                    else:
+                        if testData is None:
+                            testData = temp.reshape([1,-1])
+                        else:
+                            testData = np.concatenate([testData,temp.reshape([1,-1])])
+
+                if train_test == "train":
+                    lfile = os.path.join("labels", site, "obj_"+lr+num+".npy")
+                    temp = np.load(lfile).astype(np.int8)
+                    if trainLabel is None:
+                        trainLabel = temp[:len(os.listdir(path4))]
+                    else: 
+                        trainLabel = np.append(trainLabel, temp[:len(os.listdir(path4))])
+                    print(trainData.shape, trainLabel.shape)
+                else:
+                    if site == "lab":
+                        lfile = os.path.join("labels", site, "obj_"+lr+str(int(num)+4)+".npy")
+                    else:
+                        lfile = os.path.join("labels", site, "obj_"+lr+str(int(num)+3)+".npy")
+                    temp = np.load(lfile).astype(np.int8)
+                    if testLabel is None:
+                        testLabel = temp[:len(os.listdir(path4))]
+                    else: 
+                        testLabel = np.append(testLabel, temp[:len(os.listdir(path4))])
+                    print(testData.shape, testLabel.shape)
+
+b = np.zeros((testLabel.shape[0], 24))
+b[np.arange(testLabel.shape[0]), testLabel] = 1
+testLabel = b
+b = np.zeros((trainLabel.shape[0], 24))
+b[np.arange(trainLabel.shape[0]), trainLabel] = 1
+trainLabel = b
+print(testData.shape, trainData.shape, testLabel.shape, trainLabel.shape)
+np.save("testData",testData)
+np.save("trainData",trainData)
+np.save("testLabel",testLabel)
+np.save("trainLabel",trainLabel)

results/process01.py

@@ -0,0 +1,122 @@
+import numpy as np
+import os
+import tensorflow as tf
+
+env = os.environ.get('GRAPE_DATASET_DIR')
+testData = np.load(os.path.join(env,"testData.npy"))
+trainData = np.load(os.path.join(env,"trainData.npy"))
+testLabel = np.load(os.path.join(env,"testLabel.npy"))
+trainLabel = np.load(os.path.join(env,"trainLabel.npy"))
+
+learning_rate = 0.001
+training_iters = 200
+
+n_input = 45*80*3
+n_classes = 24
+dropout = 0.8
+
+x = tf.placeholder(tf.float32, [None, n_input])
+y = tf.placeholder(tf.float32, [None, n_classes])
+keep_prob = tf.placeholder(tf.float32)
+
+def conv2d(name, l_input, w, b):
+    return tf.nn.relu(tf.nn.bias_add(tf.nn.conv2d(l_input, w, strides=[1, 1, 1, 1], padding='SAME'),b), name=name)
+
+def max_pool(name, l_input, k):
+    return tf.nn.max_pool(l_input, ksize=[1, k, k, 1], strides=[1, k, k, 1], padding='SAME', name=name)
+
+def norm(name, l_input, lsize=4):
+    return tf.nn.lrn(l_input, lsize, bias=1.0, alpha=0.001 / 9.0, beta=0.75, name=name)
+
+def alex_net(_X, _weights, _biases, _dropout):
+    _X = tf.reshape(_X, shape=[-1, 45, 80, 3])
+
+    conv1 = conv2d('conv1', _X, _weights['wc1'], _biases['bc1'])
+    pool1 = max_pool('pool1', conv1, k=2)
+    norm1 = norm('norm1', pool1, lsize=4)
+    norm1 = tf.nn.dropout(norm1, _dropout)
+
+    conv2 = conv2d('conv2', norm1, _weights['wc2'], _biases['bc2'])
+    pool2 = max_pool('pool2', conv2, k=2)
+    norm2 = norm('norm2', pool2, lsize=4)
+    norm2 = tf.nn.dropout(norm2, _dropout)
+
+    conv3 = conv2d('conv3', norm2, _weights['wc3'], _biases['bc3'])
+    pool3 = max_pool('pool3', conv3, k=2)
+    norm3 = norm('norm3', pool3, lsize=4)
+    norm3 = tf.nn.dropout(norm3, _dropout)
+
+    dense1 = tf.reshape(norm3, [-1, _weights['wd1'].get_shape().as_list()[0]])
+    dense1 = tf.nn.relu(tf.matmul(dense1, _weights['wd1']) + _biases['bd1'], name='fc1')
+
+    dense2 = tf.nn.relu(tf.matmul(dense1, _weights['wd2']) + _biases['bd2'], name='fc2')
+
+    out = tf.matmul(dense2, _weights['out']) + _biases['out']
+    return out
+
+weights = {
+    'wc1': tf.Variable(tf.random_normal([3, 3, 3, 64])),
+    'wc2': tf.Variable(tf.random_normal([3, 3, 64, 128])),
+    'wc3': tf.Variable(tf.random_normal([3, 3, 128, 256])),
+    'wd1': tf.Variable(tf.random_normal([6*10*256, 1024])),
+    'wd2': tf.Variable(tf.random_normal([1024, 1024])),
+    'out': tf.Variable(tf.random_normal([1024, 24]))
+}
+biases = {
+    'bc1': tf.Variable(tf.random_normal([64])),
+    'bc2': tf.Variable(tf.random_normal([128])),
+    'bc3': tf.Variable(tf.random_normal([256])),
+    'bd1': tf.Variable(tf.random_normal([1024])),
+    'bd2': tf.Variable(tf.random_normal([1024])),
+    'out': tf.Variable(tf.random_normal([n_classes]))
+}
+
+pred = alex_net(x, weights, biases, keep_prob)
+
+cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y, logits=pred))
+optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)
+
+correct_pred = tf.equal(tf.argmax(pred,1), tf.argmax(y,1))
+accuracy = tf.reduce_sum(tf.cast(correct_pred, tf.float32))
+
+init = tf.global_variables_initializer()
+
+N = 200
+
+with tf.Session() as sess:
+    sess.run(init)
+    step = 1
+    # Keep training until reach max iterations
+    Ist = 0
+    Ien = Ist+N
+    while step  < training_iters:
+        if Ien > trainData.shape[0]:
+            Ien = trainData.shape[0]
+        sess.run(optimizer, feed_dict={x: trainData[Ist:Ien], y: trainLabel[Ist:Ien], keep_prob: dropout})
+        Ist+=N
+        Ien+=N
+        if Ist > trainData.shape[0]:
+            Ist = 0
+            Ien = Ist+N
+            ist = 0
+            ien = ist+N
+            temp = 0
+            while ist < trainData.shape[0]:
+                if ien > trainData.shape[0]:
+                    ien = trainData.shape[0]
+                temp += sess.run(accuracy, feed_dict={x: trainData[ist:ien], y: trainLabel[ist:ien], keep_prob: 1.})
+                ist+=N
+                ien+=N
+            acc = temp/trainData.shape[0]
+            ist = 0
+            ien = ist+N
+            temp = 0
+            while ist < testData.shape[0]:
+                if ien > testData.shape[0]:
+                    ien = testData.shape[0]
+                temp += sess.run(accuracy, feed_dict={x: testData[ist:ien], y: testLabel[ist:ien], keep_prob: 1.})
+                ist+=N
+                ien+=N
+            v_acc = temp/testData.shape[0] 
+            print('\nINFO:root:Epoch[%d] Train-accuracy=%f\nINFO:root:Epoch[%d] Validation-accuracy=%f' %(step, acc, step, v_acc))
+            step += 1