Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
evaluation removed defaults, added results caceh Jul 24, 2018
helpers added eval Jul 24, 2018
preperation added eval Jul 24, 2018
viewer added eval Jul 24, 2018
README.md Update README.md Jul 24, 2018

README.md

Scene Understanding Challenge for Autonomous Navigation in Unstructured Environments

Code for working with the dataset used for the Scene Understanding Challenge for Autonomous Navigation in Unstructured Environments. For details of getting the dataset and updates see:

For using first add helpers/ to $PYTHONPATH

The code has been tested on python 3.6.4

Dataset Structure

The structure is similar to the cityscapes dataset. That is:

  • gtFine/{split}/{drive_no}/{6 digit img_id}_gtFine_polygons.json for ground truths
  • leftImg8bit/{split}/{drive_no}/{6 digit img_id}_leftImg8bit.png for image frames

Furthermore for training, label masks needs to be generated as described bellow resulting in the following files:

  • gtFine/{split}/{drive_no}/{6 digit img_id}_gtFine_labellevel3Ids.png
  • gtFine/{split}/{drive_no}/{6 digit img_id}_gtFine_instancelevel3Ids.png

Labels

See helpers/anue_labels.py

Generate Label Masks (for training/evaluation)

python preperation/createLabels.py --datadir $ANUE --id-type $IDTYPE --color [True|False] --instance [True|False] --num-workers $C
  • ANUE is the path to the AutoNUE dataset
  • IDTYPE can be id, csId, csTrainId, level3Id, level2Id, level1Id.
  • color True generates the color masks
  • instance True generates the instance masks with the id given by IDTYPE
  • C is the number of threads to run in parallel

For the semantic segmentation challenge, masks should be generated using IDTYPE of level3Id and used for training models (similar to trainId in cityscapes). This can be done by the command:

python preperation/createLabels.py --datadir $ANUE --id-type level3Id --num-workers $C

For the instance segmentation challenge, instance masks should be generated by the following comand:

python preperation/createLabels.py --datadir $ANUE --id-type id --num-workers $C

The generated files:

  • _gtFine_labelLevel3Ids.png will be used for semantic segmentation
  • _gtFine_instanceids.png will be used for instance segmentation

Viewer

First generate label masks as described above. To view the ground truths / prediction masks at different levels of heirarchy use:

python viewer/viewer.py ---datadir $ANUE
  • ANUE has the folder path to the dataset or prediction masks with similar file/folder structure as dataset.

TODO: Make the color map more sensible.

Evaluation

Semantic Segmentation

First generate labels masks with level3Ids as described before. Then

python evaluate/evaluate_mIoU.py --gts $GT  --preds $PRED  --num-workers $C
  • GT is the folder path of ground truths containing <drive_no>/<img_no>_gtFine_labellevel3Ids.png
  • PRED is the folder paths of predictions with the same folder structure and file names.
  • C is the number of threads to run in parallel

Instance Segmentation

First generate instance label masks with ID_TYPE=id, as described before. Then

python evaluate/evaluate_instance_segmentation.py --gts $GT  --preds $PRED 
  • GT is the folder path of ground truths containing <drive_no>/<img_no>_gtFine_labellevel3Ids.png
  • PRED is the folder paths of predictions with the same folder structure and file names. The format for predictions is the same as the cityscapes dataset. That is a .txt file where each line is of the form "<instance_mask_png> ". Note that the ID_TYPE=id is used by this evaluation code.
  • C is the number of threads to run in parallel

Work in Progress

  • mIoUs at level2 and level1
  • viewer tool for masks

Acknowledgement

Some of the code was adapted from the cityscapes code at: https://github.com/mcordts/cityscapesScripts/