(WIP) Eggs + Machine learning = Egg Cam
Issues:
Decided to give in and try edge impluse. :/
How to classify socks using a Raspberry Pi, Edge Impulse, and balena
messed up and added Lowercase and Uppercase Classes
Fixed
training
half way done
woop woop 95% accurate!
i've ran through the other steps (not pictured)
found this traid off interesteing between ram and runtime speed
First image of raspbery pi cams working! and with ghetto egg conveyor belt ™️!
second picture
getting a training set form the egg conveyor belt
.....
....
Retrying the edgeimpulse + balena cam
Now training the egg classification model
Unfortunitly eggs are much harder to classify 🥺
I only got 71 percent accurancy
Looks like some of the images had had the led light used for candling bleeding over the edge of the egg... :/
These kind of images will cause the model to incorrectly guess
What did i learn today.
1 classifying candled eggs is hard. even for humans!
2 getting good training data is really hard also
3 there are alot of verables that infulces detection
- setup of camera
- lighting of scene
- getting bad data in training set
- having the egg not lined up to cover the led light
What i could i improve
- lighting shroud around light
- bigger training dataset with at least 100 of each example photos with no lens flaired eggs
- focusing the rpi cam v1 lens, or adjusting distance to its within the focal plane
- use more camera angle at once
- retrain using Auto-balance dataset
- test model with non training set eggs its never seen before
- get help figuring out why edgeimpluse + balena cam repo doesnt work