Ask Your Question

How to train soft cascade opencv_trainsoftcascade.exe

asked 2013-08-13 03:09:39 -0500

nithin gravatar image

can anyone tel how to train and generate xml using soft cascade

edit retag flag offensive close merge delete

1 answer

Sort by ยป oldest newest most voted

answered 2013-08-13 03:25:49 -0500

updated 2013-08-13 03:27:08 -0500

The softcascade is part of the master branch, which isn't complete, stable, documented and such. Basically you are on your own on stuff that is on the master branch. However, you can look at the master docs and look at the softcascade training.

However, there is the well documented cascade classifier training, which is the current stable version of a cascade classification using boost. If this is what you need, please look at:

  1. Cascade classification documentation
  2. Cascade classification training documentation
  3. Cascade classification example
edit flag offensive delete link more


I want to implement soft cascade based on "Lubomir Bourdev and Jonathan Brandt. Robust Object Detection Via Soft Cascade. IEEE CVPR, 2005" using opencv. I want to train it using my set of own features rather than using haar or lbp. I downloaded the master branch files of opencv which have softcascade and i used cmake to generate binaries. i see binaries such as opencv_perf_softcascade.exe, opencv_test_softcascade.exe & opencv_trainsoftcascade.exe.

Now i need to train and generate the xml. plz let me know is the training procedure for cascade and soft cascade is same or different. if different then a tutorial would be much helpful.

nithin gravatar imagenithin ( 2013-08-13 04:17:35 -0500 )edit

How to write the softcascade training config file??

nithin gravatar imagenithin ( 2013-08-13 04:20:22 -0500 )edit

This is exactly what I want to point out to you, there is no tutorial on the softcascade training as far as I know. However, looking behind the idea of having a single stage but many weak classifiers, I think you can reach the same result using the cascade classification, just adapting the correct parameters. So something is making me think that the basics will indeed be the same.

Using your own features will require you to dig in the code and make sure that you add an interface for pushing own features.

StevenPuttemans gravatar imageStevenPuttemans ( 2013-08-13 04:27:32 -0500 )edit

I figured out how to detect the any object with the given xml of soft cascade. if anyone knows how make configuration file for soft cascade train, then let me know. thank you in advance.

nithin gravatar imagenithin ( 2013-08-19 10:54:54 -0500 )edit

I am guessing that the actual training can be done by using the traincascade algorithm and tweaking the parameters like number of stages = 1 and then a giant number for the number of weak classifiers for that single stage. However, to know the value, this is trial and error. I think thats what changed with the softcascade approach.

StevenPuttemans gravatar imageStevenPuttemans ( 2013-08-19 16:00:26 -0500 )edit
Login/Signup to Answer

Question Tools

1 follower


Asked: 2013-08-13 03:09:39 -0500

Seen: 1,225 times

Last updated: Aug 13 '13