Re: Bag of Binary Words VS RTABMap

Posted by matlabbe on
URL: http://official-rtab-map-forum.206.s1.nabble.com/Bag-of-Binary-Words-VS-RTABMap-tp2335p2348.html

I did a longer example with CityCentre dataset (FAB-MAP dataset). Here DBoW2 finds a lot of wrong loop closures before image 680 (around where the second traversal begins), example:
...
Adding image 492: ./resources/images/0493.jpg... 
- Loop found with image 151!

Adding image 493: ./resources/images/0494.jpg... 
- Loop found with image 133!
...

Execution time:
 - Feature computation: 97.8672 ms/image
 - Loop detection: 18.0362 ms/image

With RTAB-Map using the same command line as in the previous post, results are very similar to the paper in terms of Precision/Recall (no wrong loop closures before image 680), but timings are different (faster computer and "Kp/IncrementalFlann=true"):



$ showlogs;

Loading log files...

meanTime =  103.82
maxTime =  1.2797
maxDict =  305987
maxWM =  1206
LTM =  1237
meanWM =  590.07
meanDict =    1.5500e+05
meanWordsPerSign =  399.42

Note the max time (1.3 sec). It is where the vocabulary structure needs to re-updated (kd-tree balancing). We see better on this plot (the 5 peaks show where the structure is updated, only when the vocabulary doubles in size):