EgoLanes Dataset Curation - BDD100K

[m-zain-khawaja]

Description:

Parse the BDD100K lane detection dataset to create a dataset comprising input images in PNG format and lane masks as the ground truth. Link to academic paper

• Please create a script called process_bdd100k.py in this folder which is responsible for creating the ground truth and saving the data.
• Please also create a README.md file describing the dataset and script details, in the same folder reference in the point above
• The ground truth lane information should be stored as a list of lanes, where each lane list comprises a list of keypoints, defining the lane line - in a JSON format.
• Please also classify which lanes are the ego lanes - we need to know the ego left lane, and we need to know the ego right lane - so there should be 3 classes in total (ego left, ego right, and other)
• Please ensure that keypoints are stored in relative coordinates, where the top left-most corner of the image is 0,0 and the bottom right-most corner of the image is 1,1, and all other coordinates are floating point values in the range (0,1) for x,y directions
• Please also save a binary lane mask for all lanes (lane pixels are 255 - white, and background pixels are 0 - black)
• Please ensure ground truth images, and binary lane mask are stored in PNG format
• Please also save a lane visualization mask in which the lane lines are visualized by being drawn upon the input RGB image for data auditing purposes.

Data Summary:

• RGB image in PNG Format
• Lane lines list with lane keypoints in JSON Format
• Binary Lane Mask (black and white)
• Lane Visualization Mask (not used during training, only for data auditing purposes)

Example of same task on a different dataset:

Please review this work to see an example of dataset parsing for another dataset: #40

[m-zain-khawaja]

@aadarshkt - please use the egolanes-dataset-curation---bdd100k branch

[aadarshkt]

Update - Ground truth information is saved in JSON file format with normalized keypoints, binary masks which was already available in the dataset is saved in Egolanes/create_lane/BDD100k/output/binary_masks. I have attached images for reference. Code can be found at Link to git diff

Next task - To classify lanes as Ego Left Lane, Ego right lane, and other lanes.

[aadarshkt]

Update -

1. Changed the format of the JSON to match reference format.
2. First version of annotated grouth truth images with lanes classified as {blue, green, yellow} -> {right, left, other_lanes}.

JSON

Annotated Images

Wrongly Annotated Images

In progress -

1. Cropping of image to match reference crop dimensions.
2. Finding reasons for wrongly annotated images.
3. Matching images names format with reference JSON format.