1.3.2.190214

Author: LatentLinker

AutoAi/data/dataset.py

@@ -0,0 +1,54 @@
+from utitls.imagefolder_splitter import ImageFolderSplitter
+import os, csv, cv2
+
+RESIZE = 128
+
+IMG_DIR = '/home/tian/Desktop/spiders/design/design/spiders/image'
+# Folder for storing testing images
+RESIZE_TRAIN_IMG_DIR = './data/resize/train'
+RESIZE_TEST_IMG_DIR = './data/resize/test'
+
+# Path to generate csv file
+TRAIN_CSV_DIR = './data/train_labels.csv'
+TEST_CSV_DIR = './data/test_labels.csv'
+
+
+# write csv
+def write_csv(img_dir, csv_dir):
+    list = []
+    list.append(['File Name', 'Label'])
+    for file_name in os.listdir(img_dir):
+        for img in os.listdir("%s/%s" % (img_dir, file_name)):
+            item = [file_name + "/" + img, file_name]
+            list.append(item)
+    f = open(csv_dir, 'w')
+    writer = csv.writer(f)
+    writer.writerows(list)
+
+
+# resize images
+def resize_img(path, data):
+    for i, img_file in enumerate(data[0]):
+        cls_name = data[1][i]
+        try:
+            img = cv2.imread(img_file)
+            img = cv2.resize(img, (RESIZE, RESIZE), interpolation=cv2.INTER_LINEAR)
+        except:
+            print(img_file)
+            continue
+        img_name = img_file[img_file.rfind('/') + 1:]
+        if os.path.exists("%s/%s" % (path, cls_name)):
+            cv2.imwrite("%s/%s/%s" % (path, cls_name, img_name), img)
+        else:
+            os.makedirs("%s/%s" % (path, cls_name))
+            cv2.imwrite("%s/%s/%s" % (path, cls_name, img_name), img)
+
+
+if __name__ == '__main__':
+    splitter = ImageFolderSplitter(IMG_DIR)
+    # print("Resize images...")
+    resize_img(RESIZE_TRAIN_IMG_DIR, splitter.getTrainingDataset())
+    resize_img(RESIZE_TEST_IMG_DIR, splitter.getValidationDataset())
+    # print("write csv...")
+    write_csv(RESIZE_TRAIN_IMG_DIR, TRAIN_CSV_DIR)
+    write_csv(RESIZE_TEST_IMG_DIR, TEST_CSV_DIR)

AutoAi/test_data.py renamed to AutoAi/data/test_data.py

@@ -1,40 +1,12 @@
-import os, csv, cv2
 from autokeras.image.image_supervised import load_image_dataset, ImageClassifier
 from keras.models import load_model
 from keras.utils import plot_model
 from keras.preprocessing.image import load_img, img_to_array
 import numpy as np
-from imagefolder_splitter import ImageFolderSplitter
+import cv2
 from autokeras.constant import Constant  # 微调参数设置  如batch_size  Image_size 等
 
 
-# write csv
-def write_csv(img_dir, csv_dir):
-    list = []
-    list.append(['File Name', 'Label'])
-    for file_name in os.listdir(img_dir):
-        for img in os.listdir("%s/%s" % (img_dir, file_name)):
-            item = [file_name + "/" + img, file_name]
-            list.append(item)
-    f = open(csv_dir, 'w')
-    writer = csv.writer(f)
-    writer.writerows(list)
-
-
-# resize images
-def resize_img(path, data):
-    for i, img_file in enumerate(data[0]):
-        cls_name = data[1][i]
-        img = cv2.imread(img_file)
-        img = cv2.resize(img, (RESIZE, RESIZE), interpolation=cv2.INTER_LINEAR)
-        img_name = img_file[img_file.rfind('/') + 1:]
-        if os.path.exists("%s/%s" % (path, cls_name)):
-            cv2.imwrite("%s/%s/%s" % (path, cls_name, img_name), img)
-        else:
-            os.makedirs("%s/%s" % (path, cls_name))
-            cv2.imwrite("%s/%s/%s" % (path, cls_name, img_name), img)
-
-
 def train_autokeras(RESIZE_TRAIN_IMG_DIR, RESIZE_TEST_IMG_DIR, TRAIN_CSV_DIR, TEST_CSV_DIR, TIME):
     # Load images
     train_data, train_labels = load_image_dataset(csv_file_path=TRAIN_CSV_DIR, images_path=RESIZE_TRAIN_IMG_DIR)  # 加载数据
@@ -73,7 +45,6 @@ def predict(MODEL_DIR, PREDICT_IMG_PATH, RESIZE):
 
 if __name__ == "__main__":
     # Folder for storing training images
-    IMG_DIR = '/home/tian/Desktop/spiders/design/design/spiders/image'
     # Folder for storing testing images
     RESIZE_TRAIN_IMG_DIR = './data/resize/train'
     RESIZE_TEST_IMG_DIR = './data/resize/test'
@@ -93,13 +64,6 @@ def predict(MODEL_DIR, PREDICT_IMG_PATH, RESIZE):
     RESIZE = 128
     # Set the training time, this is half an hour
     TIME = 0.5 * 60 * 60  # 训练时间  半小时
-    splitter = ImageFolderSplitter(IMG_DIR)
-    # print("Resize images...")
-    # resize_img(RESIZE_TRAIN_IMG_DIR, splitter.getTrainingDataset())
-    # resize_img(RESIZE_TEST_IMG_DIR, splitter.getValidationDataset())
-    # print("write csv...")
-    # write_csv(RESIZE_TRAIN_IMG_DIR, TRAIN_CSV_DIR)
-    # write_csv(RESIZE_TEST_IMG_DIR, TEST_CSV_DIR)
     print("============Load...=================")
-    # train_autokeras(RESIZE_TRAIN_IMG_DIR, RESIZE_TEST_IMG_DIR, TRAIN_CSV_DIR, TEST_CSV_DIR, TIME)
-    predict(MODEL_DIR, PREDICT_IMG_PATH, RESIZE)
+    # train_autokeras(RESIZE_TRAIN_IMG_DIR, RESIZE_TEST_IMG_DIR, TRAIN_CSV_DIR, TEST_CSV_DIR, TIME)  # 训练
+    predict(MODEL_DIR, PREDICT_IMG_PATH, RESIZE)  # 识别

AutoAi/train.py renamed to AutoAi/main.py

@@ -8,8 +8,8 @@ class Config(object):
     image_size = 96  # 图片尺寸
     batch_size = 32
     max_epoch = 4000
-    lr1 = 0.001  # 2e-4  # 生成器的学习率 2.7*2-4
-    lr2 = 0.001  # 2e-4  # 判别器的学习率
+    lr1 = 2e-4  # 2e-4  # 生成器的学习率 2.7*2-4
+    lr2 = 2e-4  # 2e-4  # 判别器的学习率
     beta1 = 0.5  # Adam优化器的beta1参数
     use_gpu = True  # 是否使用GPU
     nz = 100  # 噪声维度

AutoAi/utitls/__init__.py

@@ -111,7 +111,7 @@ def train(**kwargs):
 
 def generate(**kwargs):
     """
-    随机生成动漫头像，并根据netd的分数选择较好的
+    随机生成图像，并根据netd的分数选择较好的
     """
     with t.no_grad():
         opt._parse(kwargs)

AutoAi/get_classes.py renamed to AutoAi/utitls/get_classes.py

@@ -8,11 +8,11 @@ class DefaultConfig(object):
     env = 'opalus_recognltion'  # visdom 环境
     vis_port = 8097  # visdom 端口
     image_size = 224  # 图片尺寸
-    model = 'AlexNet1'  # 使用的模型，名字必须与models/__init__.py中的名字一致
+    model = 'ResNet152'  # 使用的模型，名字必须与models/__init__.py中的名字一致
 
     data_root = "/home/tian/Desktop/spiders/design/design/spiders/image"  # 数据集存放路径
     load_model_path = None  # 加载训练的模型的路径，为None代表不加载
-    # load_model_path = './checkpoint/AlexNet1_0130_12-21-31.pth.tar'
+    # load_model_path = './checkpoint/AlexNet1_0214_10-06-50.pth.tar'
 
     batch_size = 16  # 每批训练数据的个数,显存不足,适当减少
     use_gpu = True  # user GPU or not
@@ -25,7 +25,7 @@ class DefaultConfig(object):
     # pretrained = False  # 不加载预训练
     pretrained = True  # 加载预训练模型
 
-    max_epoch = 10  # 学习次数
+    max_epoch = 20  # 学习次数
     lr = 0.001  # initial learning rate
     lr_decay = 0.5  # when val_loss increase, lr = lr*lr_decay
     weight_decay = 0e-5  # 损失函数

AutoAi/imagefolder_splitter.py renamed to AutoAi/utitls/imagefolder_splitter.py

@@ -113,7 +113,7 @@ def train(**kwargs):
                                        model_name=opt.model,
                                        total=len(train_dataloader))
         for ii, (data, labels) in enumerate(train_dataloader):
-            train_progressor.current = ii
+            train_progressor.current = ii+1
             # train model
             input = data.to(opt.device)
             target = labels.to(opt.device)
@@ -139,6 +139,7 @@ def train(**kwargs):
             train_progressor()
         # train_progressor.done()  # 保存训练结果为txt
         # validate and visualize
+        print('')
         valid_loss = val(model, epoch, criterion, val_dataloader)  # 校验模型
         is_best = valid_loss[1] > best_precision  # 精确度比较，如果此次比上次大　　保存模型
         best_precision = max(valid_loss[1], best_precision)
@@ -174,7 +175,7 @@ def val(model, epoch, criterion, dataloader):
         val_progressor = ProgressBar(mode="Val  ", epoch=epoch, total_epoch=opt.max_epoch, model_name=opt.model,
                                      total=len(dataloader))
         for ii, (data, labels) in enumerate(dataloader):
-            val_progressor.current = ii
+            val_progressor.current = ii+1
             input = data.to(opt.device)
             labels = labels.to(opt.device)
             score = model(input)
@@ -187,6 +188,7 @@ def val(model, epoch, criterion, dataloader):
             val_progressor.current_loss = losses.avg
             val_progressor.current_top1 = top1.avg
             val_progressor()
+        print('')
         # val_progressor.done() # 保存校验结果为txt
         return [losses.avg, top1.avg]
 

Generate_image/config.py

@@ -46,4 +46,4 @@ def save_checkpoint(state):
     prefix = './checkpoint/' + opt.model + '_'
     filename = time.strftime(prefix + '%m%d_%H-%M-%S.pth.tar')
     torch.save(state, filename)
-    print("Get Better top1 : %s saving weights to %s" % (state["best_precision"], filename))  # 打印精确度
+    # print("Get Better top1 : %s saving weights to %s" % (state["best_precision"], filename))  # 打印精确度