# CTE-DL4CV Assignment 1 — Data 来源：https://www.kaggle.com/competitions/dl4cv1/data 本地 Meta Kaggle 快照；不是本次实时页面。 # Dataset Description

This is a simple Object Localisation task to detect the bounding boxes around an object in an image

Your task is to regress over the four parameters and output the 4 co-ordinates of the bounding box

The data of the images to be used for training and testing can be found here: https://drive.google.com/open?id=1FbyOAkNoR0mZsaj6wj93UYtmL5zOSCq9 The evaluation metric used for this assignment is DICE

Simply put, the Dice Coefficient is 2 * the Area of Overlap divided by the total number of pixels in both images the code for which is :

def dice_coef(y_true, y_pred, smooth=1):

  intersection = K.sum(y_true * y_pred, axis=[1,2,3])

  union = K.sum(y_true, axis=[1,2,3]) + K.sum(y_pred, axis=[1,2,3])

  dice = K.mean((2. * intersection + smooth)/(union + smooth), axis=0)

  return dice

Please note that you will need to convert your bounding boxes into a run-length encoding for submission

What is Run length encoding?

In order to reduce the submission file size, our metric uses run-length encoding on the pixel values. Instead of submitting an exhaustive list of indices for your segmentation, you will submit pairs of values that contain a start position and a run length. E.g. '1 3' implies starting at pixel 1 and running a total of 3 pixels (1,2,3). The competition format requires a space delimited list of pairs. For example, '1 3 10 5' implies pixels 1,2,3,10,11,12,13,14 are to be included in the mask. The pixels are one-indexed and numbered from top to bottom, then left to right: 1 is pixel (1,1), 2 is pixel (2,1), etc.

The code for converting a dataframe with bounding box co-ordinates into Run-length encoding is present in the notebooks section. Please convert into the submission format before submitting.

File descriptions

Data fields