Can you improve a classifier after you load it onto a new image?

I was wondering if it were possible to improve a classifier after I load it into a new image.

I used image A to create a classifier. (It classified everything nicely.)
Then, I loaded the classifier to image B.

However, in image B, the classifier did not classify everything perfectly, so I tried classifying some more cells to improve it and hit “Train classifier”, but the whole classification changed.
I’m assuming by doing this, I just nullified the loaded classifier and it just started a new classifier?

Thank you in advance!

Sorry, didn’t realize that was Weka, just saw the alert on segmentation :slight_smile:

Hi!

Please refer to a similar question here. According to the answer given by the creator, to improve the classifier by using different images (especially if they aren’t of the same size and you can’t make a stack of images for training), you will need to use the “Save Data” option which will save feature vectors from your training in an ARRF file. You can then load this information by using the “Load Data” option to load the saved ARRF file when classifying a new image. This will update the classifier while ensuring that data from both training instances is used in the process.

Hope this was helpful!

4 Likes

Sorry for the late replay. Indeed, what @mshah mentioned is correct. Most of the Weka classifiers are not re-trainable, so you would need to re-train them from scratch by reusing the previous images’ information.

2 Likes