loading
 
Training the avm object recognition module with a list of images
Pranav Lal from India  [9 posts]
11 year
Hi all,
Can I train the AVM object recognition module on a series of images? I have a number of images of objects, stored in a directory that I need the module to recognise. I have checked the options and do not see an option to train the module but do not see anything that will do this.

I could be missing the obvious though.

Pranav
EDV  [328 posts] 11 year
Take a look into this thread:
http://www.roborealm.com/forum/index.php?thread_id=4613
EDV  [328 posts] 11 year
You could also find out more details here:
http://www.roborealm.com/forum/index.php?thread_id=4639#
Pranav Lal from India  [9 posts] 11 year
Hi EDV,

Thanks for the links. I am a tad confused. I see that I can use the API to specify object coordinates and get the module to learn that way.

My situation is as follows.
1. I have a folder that has images of different objects such as table.jpg, apple.jpg, orange.jpg.
2. I want to be aple to ask roborealm to learn these objects.
3. Finally, roborealm should speak the name of the object if it is detected in front of the camera.

Point 3 is already taken care off. There is a nice example of roborealm speaking names assigned to faces using the avm module. My problem is in point 2. I have tried clicking on the learn object button. I guess I am confused about the function of that button. Does that button only trigger learning from the camera?
EDV  [328 posts] 11 year
Please note that the object images should be printed on the paper because AVM learns anything that is placed in interest area (red rectangle) during training (object and background too). So if background of object could change after training then further recognition would not work. Also AVM algorithm do not use color for recognition (only grayscale images). So the different colored balls cannot properly be recognized by AVM.

>> Does that button only trigger learning from the camera?

Writing to the search tree of AVM is possible with helping of API.

You should just switch NV_WR_OBJ_IDX variable from -1 into desired value of object index and it would make single write of one image (that described by NV_WR_OBJ_RECT_X, NV_WR_OBJ_RECT_Y, NV_WR_OBJ_RECT_W, NV_WR_OBJ_RECT_H variables) to the search tree of AVM.
Pranav Lal from India  [9 posts] 11 year
Hi EDP,

Many thanks. I am thinking that the original object recognition module may work betterfor me. I will now do some more eperiments.

Many of the obhjects I need to ecgnize are faes. Yes, there is a face detection module but I see no way to get the name of the face that has been detected hence my use of the object recognition and AVM navigator modules.

Pranav
EDV  [328 posts] 11 year
As you probably know the AVM algorithm also can recognize faces pretty well:
http://www.youtube.com/watch?v=4uywp5TNrZk

So you could try also to use it within your task.
Pranav Lal from India  [9 posts] 11 year
Hi EDV,
Yes, I know that the AVM module can handle face recognition. That is why I bought it. My datanbase of faces consists of images taken using a digital camera. The background varies. I am moving away from the AVM module because I don't seem to be able to give the module a list of faces to train. This appears to  be possible using the object recognition module.

Pranav

This forum thread has been closed due to inactivity (more than 4 months) or number of replies (more than 50 messages). Please start a New Post and enter a new forum thread with the appropriate title.

 New Post   Forum Index