Orb_slam2: What is Vocabulary/ORBvoc.txt for?

[jiapei100]

Sorry for my entry-level question, but what is Vocabulary/ORBvoc.txt for?
Can I just ignore this parameter during testing my own camera?

Cheers

[chwimmer]

Very basic question. But your answers is in here: Bags of Binary Words for Fast Recognition in Image Sequences

So no you cant ignore it.

[jiapei100]

@chwimmer
Thank you chwimmer, for your very prompt reply. So, Vocabularly/ORBvoc.txt cannot be ignored. But, when I use my own camera or captured images to start a fresh testing, should I use the same Vocabulary/ORBvoc.txt? Or, should I use my own Vocabulary/ORBvoc.txt?

Thanks...

[chwimmer]

Just use the same Vocabulary because its taken from a huge sets of data and works pretty good.
I recommend to use this pull request because the Vocabulary is loaded faster but it also works with the ORBvoc.txt

[jiapei100]

@chwimmer
Ok, make sense.... Current Vocabulary/ORBvoc.txt is trained out from a huge database, which is more or less a kind of generic ... I wonder how I can train my own Vocabulary/ORBvoc.txt? Any tutorial?

• How to collect sample data?
• How to train? which command?
• Anything else that we should pay attention to? etc.

Cheers

[rikpires]

Yes you could train your own data set. Please refer to https://github.com/dorian3d/DBoW2.
Make sure to collect enough sample data.

[qianwenjun0801]

@jiapei100 you should save lots of msg as voc.txt/bin saved when you load the camera and run ORB_SLAM2, and then train these msg.., the program is not very difficult, just follow the algorithm which can been seen in "DBoW2/TemplatedVocabulary.h "

[Fatmoham]

@raulmur ,@chwimmer, @rikpires, @qianwenjun0801
As mentioned in this paper "Fast Relocalisation and Loop Closing in Keyframe-Based SLAM" you created vocabulary using the data-set Bovisa 2008-09-01. This Vocabulary is taken from a huge sets of data and works pretty good for indoor and outdoor SLAM.
If I want to use the same program ORBSLAM2 in completely different environment rather than indoor and outdoor environment, are the ORBvoc.text still good to use? Or I need to generate my own Vocabulary for the data-sets I collected from that environment?

[arifle]

@Fatmoham
Have you already fixed it?
I am asking myself the same question but have no answer and I come to search and find this.
I don't exactly understand what the meaning of the 'large dataset' is. Does it mean that database is somehow common to most of the natural scene or it mean the database is a large one (or large enough) wrt all the experimental datasets used by ORBSLAM?