Generating the labels for Image Segmentation from multiple RLE strings of a image

(HarshaVardhanBabu) #1

Hi All,
I am working on generating a custom Image segmentation data set. I have different Run-Length Encoding (RLE) of different features for each image. I would like to create a data set from these to use them with fast.ai. How should I go about using SegmentationItemList to combine different RLE’s of a image into a single mask and use it as a label.

Thank you .

HarshaVardhanBabu

1 Like

(Cesar Calderon Muro) #2

Hi @Harsha1992, did you figure this out? Thanks!

0 Likes

(Pankaj Joshi) #3

I am struggling with the same for the latest Kaggle competition.

0 Likes

(HarshaVardhanBabu) #4

I did the following to get started:

  1. Processed the data frame containing the rles and dumped the masks in png format.
  2. Then used the fastai data bunch api add_test to load the masks

Look at my kaggle kernel where I followed this strategy.

0 Likes