Ask Your Question
0

“Incremental” [face] training

asked 2020-08-12 09:20:52 -0500

BradG gravatar image

I've been working with some face detection in OpenCV. I have a couple projects I've done - one does face detection which uses a _pre-built_ model. Some others do different things where I collect my own images and train my own models. When I do the latter, it's generally with much smaller datasets that what you'd use for face training.

On my face recognizer - many of the common faces I work with do _not_ get detected properly (due to odd properties like masks, hats, goggles, glasses, etc). So I want to re-train my own model - but grabbing the gigantic "stock" datasets, adding my images to it may take a VERY long time.

So the question is: is there a way to start with an existing model (XML file) and run the trainer in a way that would just _add_ my images to it?

edit retag flag offensive close merge delete

1 answer

Sort by » oldest newest most voted
1

answered 2020-08-12 09:46:13 -0500

berak gravatar image

updated 2020-08-12 10:05:13 -0500

is there a way to start with an existing model (XML file) and run the trainer in a way that would just _add_ my images to it?

no. to re-train it you would need the original faces (+your images) and the temporal stage cascades from the previous training. (you would chop off a few stages at the bottom, and retrain those with your refined data)

many of the common faces I work with do _not_ get detected properly

there are a lot of alternatives to using haar cascades here, most of them more robust (like cnn based ones) or faster (e.g. pico)

we have an example for retraining / finetuning the dnn based detector here

edit flag offensive delete link more
Login/Signup to Answer

Question Tools

1 follower

Stats

Asked: 2020-08-12 09:20:52 -0500

Seen: 33 times

Last updated: Aug 12