charles/monoloco
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
71af7e5701
monoloco/kitti-eval/README.md
Lorenzo a37b5c7b6c first commit
2020-08-20 11:33:19 +02:00

2.7 KiB
Raw Blame History

eval_kitti

Build Status License: CC BY-NC-SA

The eval_kitti software contains tools to evaluate object detection results using the KITTI dataset. The code is based on the KITTI object development kit.

Tools

  • evaluate_object is an improved version of the official KITTI evaluation that enables multi-class evaluation and splits of the training set for validation. It's updated according to the modifications introduced in 2017 by the KITTI authors.
  • parser is meant to provide mAP and mAOS stats from the precision-recall curves obtained with the evaluation script.
  • create_link is a helper that can be used to create a link to the results obtained with lsi-faster-rcnn.

Usage

Build evaluate_object with CMake:

mkdir build
cd build
cmake ..
make

The evaluate_object executable will be then created inside build. The following folders are also required to be placed there in order to perform the evaluation:

  • data/object/label_2, with the KITTI dataset labels.
  • lists, containing the .txt files with the train/validation splits. These files are expected to contain a list of the used image indices, one per row.
  • results, in which a subfolder should be created for every test, including a second-level data folder with the resulting .txt files to be evaluated.

evaluate_object should be called with the name of the results folder and the validation split; e.g.: ./evaluate_object leaderboard valsplit

parser needs the results folder; e.g.: ./parser.py leaderboard . Note: parser will only provide results for Car, Pedestrian and Cyclist; modify it (line 8) if you need to evaluate the rest of classes.

Copyright

This work is a derivative of The KITTI Vision Benchmark Suite by A. Geiger, P. Lenz, C. Stiller and R. Urtasun, used under CC BY-NC-SA. Consequently, code in this repository is published under the same Creative Commons Attribution-NonCommercial-ShareAlike 3.0 License. This means that you must attribute the work in the manner specified by the authors, you may not use this work for commercial purposes and if you alter, transform, or build upon this work, you may distribute the resulting work only under the same license.