文章目录
1.赛事背景
1.主办方
2.赛事介绍
1. 【赛题描述】
行为规范，即指某些特定的场景会对人物的行为做出特定的限制，比如加油站禁止吸烟，驾驶员禁止打电话，博物馆禁止拍照等。随着计算机人工智能的发展，这些禁止行为或者不文明行为都可通过基于视频监控的行为检测算法进行监控、发现以及适当时给与警告。
本赛题基于这些数据，要求选手设计基于计算机视觉的算法，识别出其中人物抽烟和打电话的行为，难点在于数据集中人物清晰度有限，而且人物的像素大小不一，请选手合理利用技巧，尝试不同的图像算法，提升抽烟打电话行为识别的准确度。
2. 【赛题任务】
初赛：选手可下载数据集的一部分，设计算法并完成模型训练，输出结果要求标出图片中人物的三种行为：抽烟，打电话，正常（即没有抽烟或者打电话的行人）；
复赛：选手可下载包含初赛数据集在内的全量数据，请选手利用增量数据优化初赛的算法，提升输出结果的准确性；较初赛不同的是，复赛数据集中有少量的既抽烟又打电话的人员图像，输出结构要求标出图片中人物的四种行为：抽烟，打电话，抽烟&打电话，正常。
决赛：参赛队伍现场答辩，确认最终获奖名单。
2.baseline
2.1 文件夹结构
- 1.初始文件结构
- 2.最终文件夹结构
2.2 demo
1. 01_train_test_split.py
"""
https://blog.csdn.net/u010420283/article/details/90142480
将包含三个类别文件夹的训练集划分为训练集+验证集，比例自定（此处20%）
"""
# coding=utf-8
import os, random, shutil
def moveFile(fileDir):
pathDir = os.listdir(fileDir) # 取图片的原始路径
filenumber = len(pathDir)
picknumber = int(filenumber * ratio) # 按照rate比例从文件夹中取一定数量图片
sample = random.sample(pathDir, picknumber) # 随机选取picknumber数量的样本图片
for name in sample:
shutil.move(os.path.join(fileDir, name), os.path.join(tarDir, name))
return
if __name__ == '__main__':
ori_path = r'C:\Users\hjz\python-project\project\03_smoking_calling\data\train' # 最开始train的文件夹路径
split_Dir = r'C:\Users\hjz\python-project\project\03_smoking_calling\data\val' # 移动到新的文件夹路径
ratio = 0.2 # 抽取比例
for firstPath in os.listdir(ori_path):
fileDir = os.path.join(ori_path, firstPath) # 原图片文件夹路径
tarDir = os.path.join(split_Dir, firstPath) # val下子文件夹名字
if not os.path.exists(tarDir): # 如果val下没有子文件夹，就创建
os.makedirs(tarDir)
moveFile(fileDir) # 从每个子类别开始逐个划分
2. 02_tf2_mobilev2_classes.py
# 1.导入包
import matplotlib.pyplot as plt
import numpy as np
import os
import tensorflow as tf
from tensorflow.keras.preprocessing import image_dataset_from_directory
# # 2.加载数据
train_normal = r"C:\Users\hjz\python-project\project\03_smoking_calling\data\train\normal"
train_phone = r"C:\Users\hjz\python-project\project\03_smoking_calling\data\train\phone"
train_smoke = r"C:\Users\hjz\python-project\project\03_smoking_calling\data\train\smoke"
train_dir = r"C:\Users\hjz\python-project\project\03_smoking_calling\data\train"
val_dir = r"C:\Users\hjz\python-project\project\03_smoking_calling\data\val"#验证集用01_train_test_splict.py生成
val_normal = r"C:\Users\hjz\python-project\project\03_smoking_calling\data\val\normal"
val_phone = r"C:\Users\hjz\python-project\project\03_smoking_calling\data\val\phone"
val_smoke = r"C:\Users\hjz\python-project\project\03_smoking_calling\data\val\smoke"
test_dir = r"C:\Users\hjz\python-project\project\03_smoking_calling\data\test"#测试集是需要我们预测结果
train_normal_num = len(os.listdir(train_normal))
train_phone_num = len(os.listdir(train_phone))
train_smoke_num = len(os.listdir(train_smoke))
val_normal_num = len(os.listdir(val_normal))
val_phone_num = len(os.listdir(val_phone))
val_smoke_num = len(os.listdir(val_smoke))
train_all = train_normal_num + train_phone_num + train_smoke_num
val_all = val_normal_num + val_phone_num + val_smoke_num
test_num = len(os.listdir(test_dir))
print("train normal number: ", train_normal_num)
print("train phone_num: ", train_phone_num)
print("train_smoke_num: ", train_smoke_num)
print("all train images: ", train_all)
print("val normal number: ", val_normal_num)
print("val phone_num: ", val_phone_num)
print("val_smoke_num: ", val_smoke_num)
print("all val images: ", val_all)
print("all test images: ", test_num)
# 3.设置超参数
batch_size = 32
epochs = 500
height = 150
width = 150
# 4. 数据预处理
# 图像数据预处理需要做以下工作：
# 从磁盘读入图像数据
# 解码图像内容并转成合适的格式
# 将其转换成浮点张量
# 将数值从0至255归一化到0至1之间的数值
# train_generator = tf.keras.preprocessing.image.ImageDataGenerator(rescale=1.0 / 255)
# test_generator = tf.keras.preprocessing.image.ImageDataGenerator(rescale=1.0 / 255)
#
# train_data_gen = train_generator.flow_from_directory(batch_size=batch_size,
# directory=train_dir,
# shuffle=True,
# target_size=(height, width),
# class_mode="categorical")
# test_data_gen = test_generator.flow_from_directory(batch_size=batch_size,
# directorst_dir,y=te
# target_size=(height,width),
# class_mode="categorical")
#
# # 5.数据图像可视化
#
# # sample_training_images, _ = next(train_data_gen)
# #
# # def plotImages(images_arr):
# # fig, axes = plt.subplots(1, 5, figsize=(20, 20))
# # axes = axes.flatten()
# # for img, ax in zip(images_arr, axes):
# # ax.imshow(img)
# # ax.axis("off")
# # plt.tight_layout()
# # plt.show()
# #
# # plotImages(sample_training_images[:5])
#
#
#
# # 6. 模型构建
# model = tf.keras.Sequential([
# tf.keras.layers.Conv2D(16, 3, padding="same", activation="relu", input_shape=(height, width, 3)),
# tf.keras.layers.MaxPooling2D(), # 默认池化核为2*2
# tf.keras.layers.Conv2D(32, 3, padding="same", activation="relu"),
# tf.keras.layers.MaxPooling2D(),
# tf.keras.layers.Conv2D(64, 3, padding="same", activation="relu"),
# tf.keras.layers.MaxPooling2D(),
# tf.keras.layers.Flatten(),
# tf.keras.layers.Dense(512, activation="relu"),
# tf.keras.layers.Dense(3)#三分类
# ])
#
# model.compile(optimizer=tf.keras.optimizers.Adam(),
# loss=tf.keras.losses.CategoricalCrossentropy(from_logits=True),
# metrics=['acc'])
#
# model.summary()
#
# # 7.训练模型
# history = model.fit_generator(train_data_gen,
# steps_per_epoch=train_all//batch_size,
# epochs=epochs,
# validation_data=test_data_gen,
# validation_steps=test_all//batch_size)
# # 下面尝试数据增强和dropout:
# 8.数据增强
image_gen_train = tf.keras.preprocessing.image.ImageDataGenerator(rescale=1. / 255,
rotation_range=45,#旋转角度
width_shift_range=.15,
height_shift_range=.15,
horizontal_flip=True,#水平翻转
zoom_range=0.5 #放大操作
)
train_data_gen = image_gen_train.flow_from_directory(batch_size=batch_size,
directory=train_dir,
shuffle=True,
target_size=(height, width),
class_mode="categorical")#sparse
image_gen_val = tf.keras.preprocessing.image.ImageDataGenerator(rescale=1./255)
val_data_gen = image_gen_val.flow_from_directory(batch_size=batch_size,
directory=val_dir,
target_size=(height,width),
class_mode="categorical")
# 10. 模型中加入Dropout
model_new = tf.keras.Sequential([
tf.keras.layers.Conv2D(16, 3, padding="same", activation="relu", input_shape=(height, width, 3)),
tf.keras.layers.MaxPooling2D(), # 默认池化核为2*2
tf.keras.layers.Dropout(0.2),
tf.keras.layers.Conv2D(32, 3, padding="same", activation="relu"),
tf.keras.layers.MaxPooling2D(),
tf.keras.layers.Conv2D(64, 3, padding="same", activation="relu"),
tf.keras.layers.MaxPooling2D(),
tf.keras.layers.Dropout(0.2),
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(512, activation="relu"),
tf.keras.layers.Dense(3,activation="softmax")# 3分类,得到softmax预测值类似【0.4，0.3，0.3】
])
#keras.optimizers.Adam(lr=0.001, beta_1=0.9, beta_2=0.999, epsilon=1e-08)
model_new.compile(optimizer="adam",
loss=tf.keras.losses.CategoricalCrossentropy(from_logits=False),#多分类损失
metrics=["accuracy"])
model_new.summary()
callback = tf.keras.callbacks.EarlyStopping(
monitor='val_accuracy', min_delta=0.0001,
patience=5
)
# 在文件名中包含 epoch (使用 `str.format`)
checkpoint_path = "saved_model/best.ckpt"
checkpoint_dir = os.path.dirname(checkpoint_path)
# 创建一个回调，每 5 个 epochs 保存模型的权重
cp_callback = tf.keras.callbacks.ModelCheckpoint(
filepath=checkpoint_path,
verbose=1,
save_weights_only=False,
save_best_only=True)
# # 使用 `checkpoint_path` 格式保存权重
# model_new.save_weights(checkpoint_path.format(epoch=0))
logdir = "./model"
tb_callback = tf.keras.callbacks.TensorBoard(log_dir=logdir), #log_dir将输出的日志保存在所要保存的路径中
log_callback = tf.keras.callbacks.CSVLogger(
"training_log", separator=',', append=False
)
history_new = model_new.fit_generator(train_data_gen,
steps_per_epoch=train_all//batch_size,
epochs=epochs,
validation_data=val_data_gen,
validation_steps=val_all//batch_size,
callbacks=[callback,cp_callback,tb_callback,log_callback]
)
# # 11. 结果可视化
# accuracy = history_new.history["accuracy"]
# val_accuracy = history_new.history["val_accuracy"]
# loss = history_new.history["loss"]
# val_loss = history_new.history["val_loss"]
#
# epochs_range = range(epochs)
#
# plt.figure(figsize=(8, 8))
# plt.subplot(1, 2, 1)
# plt.plot(epochs_range, accuracy, label="Training Accuracy")
# plt.plot(epochs_range, val_accuracy, label="Validation Accuracy")
# plt.legend()
# plt.title("Training and Validation Accuracy")
#
# plt.subplot(1, 2, 2)
# plt.plot(epochs_range, loss, label="Training Loss")
# plt.plot(epochs_range, val_loss, label="Validation Loss")
# plt.legend()
# plt.title("Training and Validation Loss")
# plt.show()
# 12 保存模型
# 12.1手动保存权重
# model_new.save_weights('./checkpoints/my_checkpoint')
# 12.2保存整个模型
model_new.save('saved_model/model_2')
print("保存新模型成功！")
# new_model = tf.keras.models.load_model('saved_model/my_model')
# # 检查其架构
# new_model.summary()
# 12.3 HDF5 格式
# '.h5' 扩展名指示应将模型保存到 HDF5。
# model.save('my_model.h5')
train normal number: 1915
train phone_num: 1768
train_smoke_num: 1238
all train images: 4921
val normal number: 478
val phone_num: 442
val_smoke_num: 309
all val images: 1229
all test images: 1500
Found 4921 images belonging to 3 classes.
Found 1229 images belonging to 3 classes.
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d (Conv2D) (None, 150, 150, 16) 448
_________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 75, 75, 16) 0
_________________________________________________________________
dropout (Dropout) (None, 75, 75, 16) 0
_________________________________________________________________
conv2d_1 (Conv2D) (None, 75, 75, 32) 4640
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 37, 37, 32) 0
_________________________________________________________________
conv2d_2 (Conv2D) (None, 37, 37, 64) 18496
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 18, 18, 64) 0
_________________________________________________________________
dropout_1 (Dropout) (None, 18, 18, 64) 0
_________________________________________________________________
flatten (Flatten) (None, 20736) 0
_________________________________________________________________
dense (Dense) (None, 512) 10617344
_________________________________________________________________
dense_1 (Dense) (None, 3) 1539
=================================================================
Total params: 10,642,467
Trainable params: 10,642,467
Non-trainable params: 0
_________________________________________________________________
Epoch 1/10
19/2460 [..............................] - ETA: 6:38 - loss: 1.6014 - accuracy: 0.3947
Epoch 5/500
2460/2460 [==============================] - ETA: 0s - loss: 0.9556 - accuracy: 0.5471
Epoch 00005: saving model to saved_model\cp-0005.ckpt
........
Process finished with exit code -1
3. 03_predict.py
""""
加载模型，预测，并生成json结果
"""
import cv2
import tensorflow as tf
import numpy as np
from tensorflow import keras
import os
import json
from keras.preprocessing import image
model = tf.keras.models.load_model('saved_model/model_2')
# batch_size=2
height = 150
width = 150
train_path = r"C:\Users\hjz\python-project\project\03_smoking_calling\02_tf2_somking_calling\data\train"
#
train_ds = tf.keras.preprocessing.image_dataset_from_directory(train_path,image_size=(height,width),)
class_names = train_ds.class_names
print(class_names)
from tqdm import tqdm
submission = []
test_base = './data/test'
test_data = []
for img in os.listdir(test_base):
pt = os.path.join(test_base, img)
new_pt = os.path.join('./data/test/', img)
test_data.append(new_pt)
# print(new_pt)
print('test_num:', len(test_data))
labels = sorted(os.listdir('./data/train'))
print(labels)
for im_pt in tqdm(test_data):
value = {"image_name": None, "category": None, "score": None}
img = keras.preprocessing.image.load_img(
im_pt, target_size=(height, width))
img_array = keras.preprocessing.image.img_to_array(img)
img_array = tf.expand_dims(img_array, 0) # Create a batch
predictions = model.predict(img_array)
score = predictions[0]
value['category'] = class_names[np.argmax(score)]
value['score'] = str(np.max(score))
value['image_name'] = im_pt.split('/')[-1]
submission.append(value)
with open('result.json', 'w') as f:
json.dump(submission, f)
print('done')
3.问题及改进
1. 模型太简单，需要换用resnet,efficientNet等网络
2. 未调参
4.修改记录
1.修改标签名字以符合递交要求；
2.tf2.1不好找资料改为tf2.3（有官网资料）
3.预测分数大部分为1，改下编译时指标：model_new.compile(optimizer="adam",
loss=tf.keras.losses.CategoricalCrossentropy(from_logits=False),#多分类损失
metrics=["CategoricalAccuracy"])#accuracy,CategoricalAccuracy"
4.线上结果约为85%；
版权声明：本文为HJZ11原创文章，遵循CC 4.0 BY-SA版权协议，转载请附上原文出处链接和本声明。