README.md

KITTI Evaultaion Toolkit

How to use

1. run test_kitti_pose.py

   cd $DAVO_ROOT
   CKPT_FILE="./pretrain-ckpt/model-1500000"
   TEST_OUTPUT_DIR="./outputs/"
   MODEL_NAME="model-1500000"
   VERSION="v1-decay100k-sharedNN-dilatedPoseNN-cnv6_128-segmask_all-se_flow-abs_flow-fc_tanh"
   for seq in {00..10..1} ; do 
       python test_kitti_pose.py \
           --test_seq=$seq \
           --output_dir=./$TEST_OUTPUT_DIR/$MODEL_NAME/ \
           --ckpt_file=$CKPT_FILE \
           --concat_img_dir=./kitti_odom-dump/ \
           --version=$VERSION
   done
   

2. Copy poses to the specific folder for the KITTI evaluation toolkit

   est_dir="./$TEST_OUTPUT_DIR/$MODEL_NAME"
   kitti_result_dir="./results/${MODEL_NAME}/"
   kitti_result_datadir="${kitti_result_dir}/data/"
   
   for seq in {00..10..1} ; do 
       # Copy est_file to kitti_benchmark directory
       est_file=${est_dir}/${seq}-pred_kitti_pose.txt
       cp ${est_file} ${kitti_result_datadir}/${seq}.txt
   done
   

3. (optional) generate KITTI evaluation tool: test_odometry_all

   cd $DAVO_ROOT/kitti_benchmark
   make
   

4. run testing

   ./test_odometry_all $MODEL_NAME 
   

5. run show_errors.py to show translational and rotational error for each sequence

   # Show all sequence results
   python show_errors.py results/$MODEL_NAME/errors/
   # Show specitic sequence results
   python show_errors.py results/$MODEL_NAME/errors/ -seq 0 1 2 10
   # Save results into file
   python show_errors.py results/$MODEL_NAME/errors/ -seq 0 1 2 10
   # Save all sequence results in ${MODEL_NAME}.csv file
   python show_errors.py -o results/$MODEL_NAME/errors/
   

Source

1. odometry benchmark: http://www.cvlibs.net/datasets/kitti/eval_odometry.php

➡️ [download toolkit]

2. pose ground truth: http://www.cvlibs.net/download.php?file=data_odometry_poses.zip

Other tools

1. python tools
2. matlab tools