PaddlePaddle百度论文复现营——视频分类入门学习笔记

PaddlePaddle百度论文复现营——视频分类入门学习笔记

1 任务与背景

视频分类任务、问题与挑战、经典数据集、深度学习相关背景知识

1.1 视频分类的意义

1.2 视频分类定义

1.3 视频分类的挑战

1.4 视频分类发展历程

1.5 视频分类数据集

1.5.1 HMDB-51

主要用于行为识别

1.5.2 UCF-101

主要用于行为识别，运动分析，是最具有影响力的视频分类数据集之一

1.5.3 Sports-1M

主要用于行为识别，运动分析，斯坦福大学提供的视频分类数据集，其运动信息丰富

1.5.4 Youtube-8M

主要用于行为识别，但由于视频特征并不完善，限制了算法设计发挥的空间

1.5.5 Kinetics

主要用于行为识别，由deepmind团队提出，是最具有影响力的视频分类数据集之一

1.5.6 其他视频分类数据集

1.6 传统视频分类方法vs深度学习视频分类方法

1.7 深度学习的优势

1.8 卷积神经网络(CNN)

1.9 循环神经网络(RNN)

1.10 循环神经网络(LSTM)

2 视频分类方法

双流网络、静态图像特征聚合、3D卷积等经典视频分类方法

2.1 视频分类方法概述

2.2 双流网络方法

2.3 TSN——双流网络方法

2.4 双流网络方法小结

2.5 静态图像特征聚合

2.6 CNN&LSTM——静态图像特征聚合

2.7 ActionVLAD——静态图像特征聚合

2.8 Attention Cluster——静态图像特征聚合

2.9 静态图像特征聚合小结

2.10 C3D——3D卷积方法

2.11 P3D——3D卷积方法

2.12 I3D——3D卷积方法

2.13 3D卷积方法小结

3 前沿进展

高效视频网络、运动增强的RGB分类、快慢信息结合网络、光流表示学习、时序金字塔网络

3.1 高效视频理解卷积网络

3.2 运动增强的RGB分类

3.3 SlowFast快慢信息结合网络

3.4 光流表示学习

3.5 时序金字塔网络

3.6 趋势展望

4 课程实践

TSN.py代码补全

import paddle.fluid as fluid
import numpy as np
from paddle.fluid.layer_helper import LayerHelper
from paddle.fluid.dygraph.nn import Conv2D, Pool2D, BatchNorm, Linear


class ConvBNLayer(fluid.dygraph.Layer):
    def __init__(self,
                 name_scope,
                 num_channels,
                 num_filters,
                 filter_size,
                 stride=1,
                 groups=1,
                 act=None):
        super(ConvBNLayer, self).__init__(name_scope)

        self._conv = Conv2D(
            num_channels=num_channels,
            num_filters=num_filters,
            filter_size=filter_size,
            stride=stride,
            padding=(filter_size - 1) // 2,
            groups=groups,
            act=None,
            bias_attr=False)

        self._batch_norm = BatchNorm(num_filters, act=act)

    def forward(self, inputs):
        y = self._conv(inputs)
        y = self._batch_norm(y)

        return y


class BottleneckBlock(fluid.dygraph.Layer):
    def __init__(self,
                 name_scope,
                 num_channels,
                 num_filters,
                 stride,
                 shortcut=True):
        super(BottleneckBlock, self).__init__(name_scope)

        self.conv0 = ConvBNLayer(
            self.full_name(),
            num_channels=num_channels,
            num_filters=num_filters,
            filter_size=1,
            act='relu')
        self.conv1 = ConvBNLayer(
            self.full_name(),
            num_channels=num_filters,
            num_filters=num_filters,
            filter_size=3,
            stride=stride,
            act='relu')
        self.conv2 = ConvBNLayer(
            self.full_name(),
            num_channels=num_filters,
            num_filters=num_filters * 4,
            filter_size=1,
            act=None)

        if not shortcut:
            self.short = ConvBNLayer(
                self.full_name(),
                num_channels=num_channels,
                num_filters=num_filters * 4,
                filter_size=1,
                stride=stride)

        self.shortcut = shortcut

        self._num_channels_out = num_filters * 4

    def forward(self, inputs):
        y = self.conv0(inputs)
        conv1 = self.conv1(y)
        conv2 = self.conv2(conv1)

        if self.shortcut:
            short = inputs
        else:
            short = self.short(inputs)

        y = fluid.layers.elementwise_add(x=short, y=conv2)

        layer_helper = LayerHelper(self.full_name(), act='relu')
        return layer_helper.append_activation(y)


class TSNResNet(fluid.dygraph.Layer):
    # 定义网络结构，代码补齐
    def __init__(self,
                 name_scope,
                 layers=50,
                 class_dim=102,
                 seg_num=10,
                 weight_devay=None):
        super(TSNResNet, self).__init__(name_scope)

        self.layers = layers
        self.seg_num = seg_num
        supported_layers = [50, 101, 152]
        assert layers in supported_layers, \
            "supported layers are {} but input layer is {}".format(supported_layers, layers)

        if layers == 50:
            depth = [3, 4, 6, 3]
        elif layers == 101:
            depth = [3, 4, 23, 3]
        elif layers == 152:
            depth = [3, 8, 36, 3]
        num_filters = [64, 128, 256, 512]

        self.conv = ConvBNLayer(
            self.full_name(),
            num_channels=3,
            num_filters=64,
            filter_size=7,
            stride=2,
            act='relu')

        self.pool2d_max = Pool2D(
            pool_size=3,
            pool_stride=2,
            pool_padding=1,
            pool_type='max')

        self.bottleneck_block_list = []
        num_channels = 64
        for block in range(len(depth)):
            shortcut = False
            for i in range(depth[block]):
                bottleneck_block = self.add_sublayer(
                    'bb_%d_%d' % (block, i),
                    BottleneckBlock(
                        self.full_name(),
                        num_channels=num_channels,
                        num_filters=num_filters[block],
                        stride=2 if i == 0 and block != 0 else 1,
                        shortcut=shortcut))
                num_channels = bottleneck_block._num_channels_out
                self.bottleneck_block_list.append(bottleneck_block)
                shortcut = True

        self.pool2d_avg = Pool2D(pool_size=7, pool_type='avg', global_pooling=True)

        import math
        stdv = 1.0 / math.sqrt(2048 * 1.0)

        self.out = Linear(input_dim=num_channels,
                          output_dim=class_dim,
                          act='softmax',
                          param_attr=fluid.param_attr.ParamAttr(initializer=fluid.initializer.Uniform(-stdv, stdv)))

    def forward(self, inputs, laber=None):
        out = fluid.layers.reshape(inputs, [-1, inputs.shape[2], inputs.shape[3], inputs.shape[4]])
        y = self.conv(out)
        y = self.pool2d_max(y)
        for bottleneck_block in self.bottleneck_block_list:
            y = bottleneck_block(y)
        y = self.pool2d_avg(y)
        out = fluid.layers.reshape(x=y, shape=[-1, self.seg_num, y.shape[1]])

        out = fluid.layers.reduce_mean(out, dim=1)
        y = self.out(out)

        if laber is not None:
            acc = fluid.layers.accuracy(input=y, label=laber)
            return y, acc
        else:
            return y


if __name__ == '__main__':
    with fluid.dygraph.guard():
        network = TSNResNet('resnet', 50)
        img = np.zeros([1, 10, 3, 224, 224]).astype('float32')
        img = fluid.dygraph.to_variable(img)
        outs = network(img).numpy()
        print(outs)