1 · Introduction

An accurate High Definition (HD) Maps with lane markings usually serves as the back-end for all commercial auto-drive vehicles for navigation. Currently, most HD maps are constructed manually by human labelers. In this challenge, we require participants to develop algorithms to extract all basic road elements from RGB image frames. The segmentation results can be directly used for HD Maps construction or updating process.

This repository contains the evaluation scripts for the landmark detection challenge of the ApolloScapes dataset. This large-scale dataset contains a diverse set of stereo video sequences recorded in street scenes from different cities, with high quality pixel-level annotations of 110 000+ frames.

2 · Dataset Download

Request to download full data: Please Send Email.

Sample data

Training data

Labels_road02.tar.gz Labels_road03.tar.gz

Testing data

3 · Dataset Structure

The folder structure of the landmark detection challenge is as follows:

{root}/{type}_{road id}/{type}/{record id}/{camera id}/{timestamp}_{camera id}{ext}

The meaning of the individual elements is:

  • root the root folder of the Apolloscapes dataset.
  • type the type/modality of data, e.g. ColorImage and 'Labels'.
  • road id an identifier specifying the road, e.g. road02.
  • record id the folder name of a subset of images. Defined by the data collection system.
  • camera id images are grouped by the cameras that capture them. In Apolloscape there are always two cameras: 'Camera 5' and 'Camera 6'.
  • timestamp the time when each image is captured.
  • ext the extension of the file. '.jpg' for RGB images and '.png' for groundtruths.

4 · Scripts

The evaluation scripts are released on github here. For lane segmentation, the codes are under /dataset-api/lane_segmentation/. The structure of the script is described as below:

  • helpers helper files which include usful information of using our evaluation files
  • evaluation the main file for validating your approach
  • thirdParty containing scripts from external libraries. We borrow some codes from Cityscapes.

Note that all files have a short description at the top. Most important files are listed as below

  • helpers/laneMarkDetection.py the main file defining the IDs of all lane classes and providing mapping between various class properties.
  • evaluation/evalPixelLevelSemanticLabeling.py script to evaluate pixel-level lane labeling results on the test set.
  • install.sh installation script of this library. Only tested for Ubuntu.

The scripts can be installed by running install.sh in the bash: sudo bash install.sh

This tool is dependent on the evaluation script from cityScape dataset, which is will be pulled recursively

5 · Evaluation

Once you want to test your method on the test set, please run your approach on the provided test images and submit your results to our ApolloScape Challenge:

For semantic labeling, we require the result format to match the format of our label images. Thus, your code should produce images where each pixel's value corresponds to a class ID as defined in laneMarkDetection.py. Note that our evaluation scripts are included in the scripts folder and can be used to test your approach. For further details regarding the submission process, please consult our website.

Run the following code for a sample evaluation:

python evaluation/evalPixelLevelSemanticLabeling.py ./test_eval_data/ ./test_eval_data/pred_list.csv ./test_eval_data/ ./test_eval_data/gt_list.csv

6 · Metric formula

We adopt the widely used mean IoU metric which is presented in cityscape metric here. For each class, given the predicted masks and ground truth of image and class , the metric for evaluation is defined as:

7 · Rules of ranking

Result benchmark will be:

Method mean iou lane name 1 lane name 2 lane name 3
Deepxxx xx xx xx xx

Our ranking will determined by the mean iou of all lane classes.

Submission of data format

  • Example dir tree of submitted zip file
├── test
│   ├── road_name_1
│   │   ├── image_name1.png
│   │   ├── image_name1.png
│   │    ...
│   ├── road_name_2
│   │   ├── image_name1.png
│   │   ├── image_name2.png
  • Example format of image_name1.png
  1. image_name1.png is a prediction label image, which should have the same name and same size as the testing image. In this image, each pixel encode the class IDs as defined in our labels description. Note that regular ID is used, not the train ID.
  2. Each pixel is encoded as uint8 format.
The dataset we released is  desensitized street view  for academic use only.