mhsszm / gslam_bowbench

Comparison of DBoW2, DBoW3, FBoW and Vocabulary of GSLAM.

GSLAM Vocabulary Benchmark

This project performs comparison between DBoW2, DBoW3, FBoW and Vocabulary of GSLAM. The load, save, transform and memory usage of them are compared.

Usage

1. Download the source code

git clone --recursive https://github.com/pi-lab/gslam_bowbench

2. Compile

cd gslam_bowbench
mkdir build
cd build
cmake ..
make

3. Run

The program need a prepared file with image path listed. You can generate a sample list file with bash:

for img in $(ls ../DBow3/utils/images/*);do echo $img;done > images.txt

Show usage with:

gslam_bowbench --help 

Run with:

gslam_bowbench -images images.txt -k=10 -level=3 -mem -feature=ORB

4. Sample result

Here is a sample result:

build> ./gslam_bowbench -images images.txt 
main.cpp:328 Memory analysis started.
main.cpp:353 ../DBow3/utils/images/image0.png found 843 keypoints.
main.cpp:353 ../DBow3/utils/images/image1.png found 908 keypoints.
main.cpp:353 ../DBow3/utils/images/image2.png found 862 keypoints.
main.cpp:353 ../DBow3/utils/images/image3.png found 896 keypoints.
----------------GSLAM------------------------
main.cpp:26 GSLAM: Creating vocabulary from image features.

main.cpp:34 Vocabulary: k = 10, L = 3, Weighting = tf-idf, Scoring = L1-norm, Number of words = 1000

main.cpp:51 GSLAM used memory 44536 bytes with 5 pieces.
main.cpp:55 Memory leak:0 bytes with 0 pieces.
----------------DBoW2------------------------
main.cpp:89 DBoW2: Creating vocabulary from image features.
main.cpp:97 Created Vocabulary: k = 10, L = 3, Weighting = tf-idf, Scoring = L1-norm, Number of words = 980
main.cpp:116 DBoW2: Saving vocabulary Vocabulary: k = 10, L = 3, Weighting = tf-idf, Scoring = L1-norm, Number of words = 980
main.cpp:130 DBoW2 used memory 245932 bytes with 1204 pieces.
main.cpp:133 Memory leak:0 bytes with 0 pieces.
----------------DBoW3------------------------
main.cpp:204 DBoW3: Creating vocabulary from image features.
main.cpp:209 Created Vocabulary: k = 10, L = 3, Weighting = tf-idf, Scoring = L1-norm, Number of words = 991
main.cpp:228 DBoW3: Saving vocabulary Vocabulary: k = 10, L = 3, Weighting = tf-idf, Scoring = L1-norm, Number of words = 991
main.cpp:242 DBoW3 used memory 248628 bytes with 1217 pieces.
main.cpp:245 Memory leak:0 bytes with 0 pieces.
----------------FBoW------------------------
main.cpp:255 FBoW: Creating vocabulary from image features.
main.cpp:263 Created 111
main.cpp:283 FBoW: Saving vocabulary ...
main.cpp:297 FBoW used memory 45464 bytes with 3 pieces.
main.cpp:300 Memory leak:0 bytes with 0 pieces.

------------------------------------  Timer report ------------------------------------
           FUNCTION                       #CALLS  MIN.T  MEAN.T  MAX.T  TOTAL 
---------------------------------------------------------------------------------------
DBoW2::load                                  1    4.9ms   4.9ms   4.9ms   4.9ms
DBoW2::save                                  1    4.2ms   4.2ms   4.2ms   4.2ms
DBoW2::train                                 1  114.0ms 114.0ms 114.0ms 114.0ms
DBoW2::transDes                           3509  617.0ns 761.0ns   3.3us   2.7ms
DBoW2::transImage                            4  994.1us   1.1ms   1.2ms   4.4ms
DBoW3::load                                  1  870.9us 870.9us 870.9us 870.9us
DBoW3::save                                  1  300.4us 300.4us 300.4us 300.4us
DBoW3::train                                 1  152.1ms 152.1ms 152.1ms 152.1ms
DBoW3::transDes                           3509  329.0ns 463.8ns   6.9us   1.6ms
DBoW3::transImage                            4  743.6us 857.5us 985.2us   3.4ms
FBoW::load                                   1   24.3us  24.3us  24.3us  24.3us
FBoW::save                                   1  217.2us 217.2us 217.2us 217.2us
FBoW::train                                  1  116.8ms 116.8ms 116.8ms 116.8ms
FBoW::transDes                            3509  530.0ns 558.4ns   1.1us   2.0ms
FBoW::transImage                             4  300.3us 352.4us 377.6us   1.4ms
GSLAM::load                                  1   38.0us  38.0us  38.0us  38.0us
GSLAM::save                                  1  179.3us 179.3us 179.3us 179.3us
GSLAM::train                                 1   72.7ms  72.7ms  72.7ms  72.7ms
GSLAM::transDes                           3509  239.0ns 301.0ns 502.0ns   1.1ms
GSLAM::transImage                            4  273.2us 291.1us 302.1us   1.2ms
--------------------------------- End of Timer report ---------------------------------