The competition leaderboard will use the Root Mean Square Error Metric (RMSE) for purpose of score calculation.

$$
RMSE = \sqrt{\frac{1}{n}\Sigma\_{i=1}^{n}{\Big(\frac{d\_i -f\_i}{\sigma\_i}\Big)^2}}
$$

This metric will be used per pixel value of each image and each image will be compared to ground truth images. These can be seen in the train set as cleaned training data. The images are all grayscale and thus all pixel values are to be calculated by finding a float value within a scale of 0 to 1. 

The output is to be submitted in a CSV as given in the sample submission. Each pixel value becomes a new row. For example, a pixel at position (50,100) of image 120 with value 0.965 will be represented as:

| Display Type | ID | Prediction |
| --- | --- | --- |
| Row Template | imgno_x_y | value |
| Real | 120_50_100 | 0.965 |


The scores will be the summation of all the RMSE values of each pixel. The lower the score, the higher the leaderboard ranking.