# OCR model evaluation
###### tags: `cv_infra`

## 1-N.E.D.

* ICDAR 2019 官方是說會把空白去除跟標點符號去除
* https://arxiv.org/pdf/1909.07145.pdf

$$
Norm = 1 - \frac{1}{N}\sum_\limits{i=1}^{N}D(p_i, a_i)/max(p_i,a_i)
$$

* $D$ is Levenshtein Distance
* https://github.com/miohtama/python-Levenshtein/blob/master/StringMatcher.py
* edit-distance
* https://github.com/belambert/edit-distance
## accuarcy

* 我們所知道的

## 之後考慮看看

### coverage

* 包含 prob, recall 所畫出來的圖、表格
* prob = min(prob_list) :star:
* prob = mean(prob_list)

### precision-recall

### AUC?