将视频数据集转换为帧序列，并划分为训练集，验证集与测试集（以HMDB51为例）

首先需要观察数据集的存储路径
HMDB51原始存储是路径+类（文件夹）+视频（avi视频）
提取帧序列我们需要转换为路径+类+视频名（文件夹）+帧序列（jpg图片）
划分数据集后存储路径为路径+训练/验证/测试（文件夹）+类（文件夹）+帧序列

提取帧序列代码：

import os
import numpy as np
from PIL import Image
import cv2
import stat
# 将avi数据集视频抽帧，转换为帧序列数据集
# 帧数间隔
frameRate=1

# videoFileName为视频的总路径，save_paths为保存路径+类
# videoFileName为E:\Master's period\Gesture dataset\HMDB51\hmdb51\brush_hair\April_09_brush_hair_u_nm_np1_ba_goo_0.avi
# save_path为E:\Master's period\Gesture dataset\HMDB51_frame\brush_hair\
def splitFrames(videoFileName,save_path):
	# 打开视频文件
    cap = cv2. VideoCapture(videoFileName) 
    temp1 = str(videoFileName.split("/")[-2])
    temp2 = str(videoFileName.split("/")[-1])
    save_dir = os.path.join(save_path,temp1,temp2)
    if not os.path.exists(save_dir):
        os.makedirs(save_dir)
    num = frameRate
    while True:
        # success 表示是否成功，data是当前帧的图像数据；.read读取一帧图像，移动到下一帧
        success, data = cap.read()
        if not success:
            break
        if num % frameRate == 0:
            data = np.array(data)[:, :, ::-1]
            im = Image.fromarray(data) # 重建图像
            im.save(save_dir + '/' +str(int(num/frameRate))+".jpg") # 保存当前帧的静态图像
        num = num + 1
    print("保存成功")
    cap.release()


# 初始文件夹
file_path=r"E:\Master's period\Gesture dataset\HMDB51\hmdb51"
os.chmod(file_path,stat.S_IRWXU)
# 移动后文件夹
save_path=r"E:\Master's period\Gesture dataset\HMDB51_frame"
# file_dirs为所有类的列表
file_dirs=os.listdir(file_path)
# 定义要保存的路径dir
origion_paths=[]
save_paths=[]
# 循环所有类，添加到初始文件夹与保存文件夹下
for path in file_dirs:
    # origion_paths为初始路径+类
    origion_paths.append(file_path + "/" + path + "/")
    # save_paths为保存路径+brush_hair序列
    # save_paths.append(save_path + "/" + path + "/")
# 在类别下获取两个，i为序列，origion_path为获取的每个路径
# 如0 E:\Master's period\Gesture dataset\HMDB51\hmdb51\brush_hair
for i,origion_path in enumerate(origion_paths):
    # 获得原始路径下的所有图片的name（默认路径下都是图片）
    # image_list为brush_hair类别下所有视频名字序列
    image_list = os.listdir(origion_path)
    image_Dir = []
    for x, y in enumerate(image_list):
        # image_Dir为E:\Master's period\Gesture dataset\HMDB51\hmdb51\brush_hair\April_09_brush_hair_u_nm_np1_ba_goo_0.avi
        image_Dir.append(os.path.join(origion_path, y))
        splitFrames(image_Dir[x],save_path)

print("all datas has been moved successfully!")

划分数据集代码：

import os
import random
import shutil

# 将帧序列按7：1：2划分为训练集，验证集与测试集

fileDir = "E:/Master's period/Gesture dataset/HMDB51_frame/"
trainDir = "E:/Master's period/Gesture dataset/HDMB51_dataset/train/"
valDir = "E:/Master's period/Gesture dataset/HDMB51_dataset/val/"
testDir = "E:/Master's period/Gesture dataset/HDMB51_dataset/test/"


def copyFile(fileDir,trainDir,valDir,testDir):
    # 取图片原始路径
    # lei为所有类别目录
    lei = os.listdir(fileDir)
    trainpath = []
    valpath = []
    testpath = []
    filepath = []
    for x in lei:
        # trainpath里存放的是E:/Master's period/Gesture dataset/HDMB51_dataset/train/brush_hair/
        trainpath.append(trainDir + x + "/")
        valpath.append(valDir + x + "/")
        testpath.append(testDir + x + "/")
        filepath.append(fileDir + x + "/")
    for i,path in enumerate(filepath):
        # path为E:/Master's period/Gesture dataset/HMDB51_frame/brush_hair/
        pathDir = os.listdir(path)
        num = len(pathDir)
        # 移动train
        picknumber1 = int(num * 0.7)
        # pathDir为总视频的list
        sample1 = random.sample(pathDir,picknumber1)
        for name in sample1:
            dile_dir = os.path.join(path, name)
            save_dir = os.path.join(trainpath[i], name)
            shutil.move(dile_dir,save_dir)
        print("train移动完成")
        # 移动val
        pathDir = os.listdir(path)
        picknumber2 = int(num * 0.1)
        sample2 = random.sample(pathDir,picknumber2)
        for name in sample2:
            dile_dir = os.path.join(path, name)
            save_dir = os.path.join(valpath[i], name)
            shutil.move(dile_dir,save_dir)
        print("val移动完成")
        # 移动test,将文件夹下剩余文件都移动（防止有残余）
        pathDir = os.listdir(path)
        # picknumber3 = int(num * 0.2)
        # sample3 = random.sample(pathDir,picknumber3)
        for name in pathDir:
            dile_dir = os.path.join(path, name)
            save_dir = os.path.join(testpath[i], name)
            shutil.move(dile_dir, save_dir)
        print("test移动完成")


if __name__ =='__main__':
    copyFile(fileDir,trainDir,valDir,testDir)