论文阅读——长难句理解和记录

1. Our technique first recovers the original 3D camera motion and ->从这里开始断句 a sparse
   set of 3D, static scene points using an off-the-shelf structure-frommotion system.

static：静态
dynamic：动态

quadratic：二次的
parabolic：抛物线形的
scalar && vector ：标量和矢量
stochastic gradient descent : 随机梯度下降
magnitude: n. 大小；量级；[地震] 震级；重要；光度
order of magnitude 数量级
signal magnitude 信号幅度

spatially-variant ：在空间上变化的。这是个形容词。
terms and notations ：术语和符号
punctuation:标点
phase :阶段
phrase ：句子

2. the third subnetwork generates relevant textual captions using as input the spatiotemporal features of the second subnetwork.
   第三子网使用第二子网的时空特征作为输入生成相关的文本字幕
   相当于用了一个倒装。
   using XXXXX as input.
   （using as input 也可以看做是强调吧）

3. As was the case with the encoder, the decoder’s architecture must be chosen according to the type of the output.
   与编码器的情况一样，必须根据输出类型选择解码器的架构。

4. then the (7) can be rewritten by left multiplication inverse R
   通过左乘R的逆，把等式（7）重写成

5. the first time step didsomething, following the rest time steps with input from the previous hidden state
   第一个time step干了啥，然后接下来的time step的输入来自前面的隐藏层。

6. The memory cell updates its hidden state by combining the previous cell state which is modulated by the forget gate and a function of the current input and the previous output, modulated by the input gate.
   input gate到底modulate谁？是这个 a function还是以前输出？
   记忆单元通过综合 以前单元的状态（由遗忘门调整）、当前输入和以前输出的函数（由输入门）

7. in our experiments, we instead consider techniques to automatically extract or predict attributes that are not directly given.
   在这个实验中，我们改为考虑自动提取或预测未直接给出的属性的技术