Cricket Bats vs Baseball Bats

Hello Fast AI Community,

I have just started watching the v2 version of course (just finished 30 minutes of lesson 1 - first video).

Validation set - 10 images
Training set - 20 images

Cricket Bat had rough value (validation set random value) of 0.8 while for Baseball Bat had a value of around 0.5.

Just thought of sharing the results of the experiment.
I assume with more training sets or better quality of the images the results would be better.

I change the batch size parameter and I was able to get better accuracy.
0 - baseball bat 1 - cricket bat

data = ImageClassifierData.from_paths(path, tfms=tfms_from_model(arch, sz), bs=bs)

Great. You could have a look at this for more info!


Nice article. That is some really good results such a small training set.