## object as point阅读笔记

### 方法初步

$Y_{x y c}=\exp \left(-\frac{\left(x-\tilde{p}_{x}\right)^{2}+\left(y-\tilde{p}_{y}\right)^{2}}{2 \sigma_{p}^{2}}\right)$

$L_{k}=\frac{-1}{N} \sum_{x y c}\left\{\begin{array}{cl} \left(1-\hat{Y}_{x y c}\right)^{\alpha} \log \left(\hat{Y}_{x y c}\right) & \text { if } Y_{x y c}=1 \\ \left(1-Y_{x y c}\right)^{\beta}\left(\hat{Y}_{x y c}\right)^{\alpha} & \\ \log \left(1-\hat{Y}_{x y c}\right) & \text { otherwise } \end{array}\right.$

$\hat{O} \in \mathcal{R}^{\frac{W}{R} \times \frac{H^{\prime}}{R} \times 2}$

$L_{o f f}=\frac{1}{N} \sum_{p}\left|\hat{O}_{\tilde{p}}-\left(\frac{p}{R}-\tilde{p}\right)\right|$

### Objects as Points

$$\left(x_{1}^{(k)}, y_{1}^{(k)}, x_{2}^{(k)}, y_{2}^{(k)}\right)$$为物体$$k$$，类别$$c_k$$的的bbx。作者使用他们的keypoint estimator $$\hat{Y}$$， 来预测所有的中心点，除此之外还会预测物体$$k$$的尺寸$$s_{k}=\left(x_{2}^{(k)}-x_{1}^{(k)}, y_{2}^{(k)}-y_{1}^{(k)}\right)$$。为了减少计算压力，作者对所有的物体种类都使用单个size predictor $$\hat{S} \in \mathcal{R}^{\frac{W}{R} \times \frac{H}{R} \times 2}$$. 同时作者使用了L1 loss，定义如下

$L_{\text {size }}=\frac{1}{N} \sum_{k=1}^{N}\left|\hat{S}_{p_{k}}-s_{k}\right|$

$L_{\text {det }}=L_{k}+\lambda_{\text {size }} L_{\text {size }}+\lambda_{\text {of } f} L_{\text {off }}$

$\begin{array}{l} \left(\hat{x}_{i}+\delta \hat{x}_{i}-\hat{w}_{i} / 2, \quad \hat{y}_{i}+\delta \hat{y}_{i}-\hat{h}_{i} / 2\right. \\ \left.\hat{x}_{i}+\delta \hat{x}_{i}+\hat{w}_{i} / 2, \quad \hat{y}_{i}+\delta \hat{y}_{i}+\hat{h}_{i} / 2\right) \end{array}$

#### 人体姿态估计

$l_{j}=(\hat{x}, \hat{y})+\hat{J}_{\hat{x} \hat{y} j} \text { for } j \in 1 \ldots k$

posted on 2021-05-26 14:05  YongjieShi  阅读(131)  评论(0编辑  收藏  举报