building_with_espdl-1-failure

building_with_espdl-1-failure

1
2
import pickle
import os
1
import tensorflow as tf
1
2
from keras.models import Sequential
from keras.layers import Conv2D, MaxPooling2D, Dense, Flatten, Dropout
1
2
3
import tensorflow as tf
print(tf.__version__)          # 打印 TensorFlow 版本
print(tf.keras.__file__)       # 打印 Keras 的实际路径

这两行是上面打印版本和实际路径的输出结果

2.17.0
/home/dahao/anaconda3/envs/lowgensim/lib/python3.10/site-packages/keras/api/_tf_keras/keras/__init__.py

这一步pkl的作用尚且未知，不懂pkl的意义是什么

1
2
3
4
5
6
7
8
9
10
11
with open('/home/dahao/code/Blogs/ESP-DL/model_development/X_test.pkl', 'rb') as file:
    X_test = pickle.load(file)

with open('/home/dahao/code/Blogs/ESP-DL/model_development/y_test.pkl', 'rb') as file:
    y_test = pickle.load(file)

with open('/home/dahao/code/Blogs/ESP-DL/model_development/X_train.pkl', 'rb') as file:
    X_train = pickle.load(file)

with open('/home/dahao/code/Blogs/ESP-DL/model_development/y_train.pkl', 'rb') as file:
    y_train = pickle.load(file)

从数据集中选了其中一个文件夹中的其中一部分数据

1
2
# 设置数据集路径
data_dir = "./leapgestrecog/leapGestRecog"
1
2
X = []
y = []
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
from PIL import Image
import numpy as np
# 读取每个手势类别文件夹中的图像文件
for gesture_folder in os.listdir(data_dir):
    gesture_folder_path = os.path.join(data_dir, gesture_folder)
    if os.path.isdir(gesture_folder_path):  # 检查是否是文件夹
        for gesture_instance in os.listdir(gesture_folder_path):
            instance_dir = os.path.join(gesture_folder_path, gesture_instance)
            if os.path.isdir(instance_dir):  # 检查是否是文件夹
                for image_file in os.listdir(instance_dir):
                    image_path = os.path.join(instance_dir, image_file)
                    # 读取图像并进行预处理
                    image = Image.open(image_path).convert("L")  # 转为灰度图像
                    image = image.resize((96, 96))  # 调整图像大小
                    X.append(np.array(image))  # 将图像添加到列表
                    y.append(gesture_instance)  # 使用手势实例作为标签
1
2
3
4
# 将列表转为 NumPy 数组
X = np.array(X)
y = np.array(y)

检验数据集形状和标签形状

1
print(f"数据集形状: {X.shape}, 标签形状: {y.shape}")
数据集形状: (20000, 96, 96), 标签形状: (20000,)

1
2
3
4
5
6
from sklearn.model_selection import train_test_split

ts = 0.3 # Percentage of images that we want to use for testing. 
X_train, X_test1, y_train, y_test1 = train_test_split(X, y, test_size=ts, random_state=42)
X_test, X_cal, y_test, y_cal = train_test_split(X_test1, y_test1, test_size=ts, random_state=42)

cnn算法部分

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
print(tf.__version__)

model = Sequential()
model.add(Conv2D(32, (5, 5), activation='relu', input_shape=(96, 96, 1))) 
model.add(MaxPooling2D((2, 2)))
model.add(Dropout(0.2))
model.add(Conv2D(64, (3, 3), activation='relu')) 
model.add(MaxPooling2D((2, 2)))
model.add(Dropout(0.2))
model.add(Conv2D(64, (3, 3), activation='relu'))
model.add(MaxPooling2D((2, 2)))
model.add(Flatten())
model.add(Dense(128, activation='relu'))
model.add(Dense(6, activation='softmax'))

model.compile(optimizer='adam',loss='sparse_categorical_crossentropy',metrics=['accuracy'])

model.summary()

这部分是cnn算法的输出结果，和原作者的相吻合

2.17.0

Model: "sequential_8"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Layer (type)                    ┃ Output Shape           ┃       Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ conv2d_24 (Conv2D)              │ (None, 92, 92, 32)     │           832 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ max_pooling2d_24 (MaxPooling2D) │ (None, 46, 46, 32)     │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dropout_16 (Dropout)            │ (None, 46, 46, 32)     │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_25 (Conv2D)              │ (None, 44, 44, 64)     │        18,496 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ max_pooling2d_25 (MaxPooling2D) │ (None, 22, 22, 64)     │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dropout_17 (Dropout)            │ (None, 22, 22, 64)     │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_26 (Conv2D)              │ (None, 20, 20, 64)     │        36,928 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ max_pooling2d_26 (MaxPooling2D) │ (None, 10, 10, 64)     │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ flatten_8 (Flatten)             │ (None, 6400)           │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_16 (Dense)                │ (None, 128)            │       819,328 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_17 (Dense)                │ (None, 6)              │           774 │
└─────────────────────────────────┴────────────────────────┴───────────────┘

 Total params: 876,358 (3.34 MB)

 Trainable params: 876,358 (3.34 MB)

 Non-trainable params: 0 (0.00 B)

尝试解决部分，未成功

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
# 添加通道维度
X_train = X_train.reshape(X_train.shape[0], 96, 96, 1)
X_test = X_test.reshape(X_test.shape[0], 96, 96, 1)

# 确保数据类型为float32
X_train = X_train.astype('float32') / 255.0
X_test = X_test.astype('float32') / 255.0

# 对标签进行编码
from keras.utils import to_categorical
y_train = to_categorical(y_train)
y_test = to_categorical(y_test)

# 训练模型
history = model.fit(X_train, y_train, epochs=5, batch_size=64, verbose=1, validation_data=(X_test, y_test))

一直显示这部分报错结果

---------------------------------------------------------------------------

ValueError                                Traceback (most recent call last)

Cell In[272], line 11
      9 # 对标签进行编码
     10 from keras.utils import to_categorical
---> 11 y_train = to_categorical(y_train)
     12 y_test = to_categorical(y_test)
     14 # 训练模型


File ~/anaconda3/envs/lowgensim/lib/python3.10/site-packages/keras/src/utils/numerical_utils.py:86, in to_categorical(x, num_classes)
     84         x = backend.numpy.reshape(x, newshape)
     85     return backend.nn.one_hot(x, num_classes)
---> 86 x = np.array(x, dtype="int64")
     87 input_shape = x.shape
     89 # Shrink the last dimension if the shape is (..., 1).


ValueError: invalid literal for int() with base 10: '01_palm'

1
type(y_train)
numpy.ndarray

今日有效学习链接：

1、本文基于并且想尝试复现的github文档：
https://github.com/alibukharai/Blogs/blob/main/ESP-DL/building_with_espdl.md
2、No module named ‘tensorflow.keras‘报错信息的解决方法：
https://blog.csdn.net/Zinnir/article/details/125999939?fromshare=blogdetail&sharetype=blogdetail&sharerId=125999939&sharerefer=PC&sharesource=m0_60571820&sharefrom=from_link
我没有用python，我用的是shell，原理相同
3、kaggle怎么下载数据集？
https://blog.csdn.net/weixin_73577120/article/details/143170847?fromshare=blogdetail&sharetype=blogdetail&sharerId=143170847&sharerefer=PC&sharesource=m0_60571820&sharefrom=from_link

原来失恋是这种感觉，也有可能我本来就从未得到过吧……一直是我在自我幻想罢了，无所谓，短时间我不会再相信爱情了……
人，总是要向前看的，找不到答案的话就找自己吧！

fyp
building_with_espdl-1-failure
https://blakehansen130.github.io/2024/11/01/Ali/
发布于
2024年11月1日
许可协议