I captured three states of device (doing POC) -
- Breadboard with switch (Label - Found)
- Breadboard without switch (Label - Not Found)
- No Breadboard (Label - Empty)
With current Image classification, I could manage to predict labels like Empty , but classifying Found or Not Found is not very accurate.
Any advice here.
It might depend on your dataset. Are you using grayscale or RGB? For Grayscale, it might be apt to capture images with the switch visible clearly. Also, is there a bias in your dataset, or are the images in each class unequal? You could tweak with the Mobilenet Version, and see if MobileNet v2 0.35 Model works well.
1 Like
@dhruvsheth let me try that,
btw do you know how I can view live classification via url “http://localhost:4911”. It seems it does not work. I just looked at instruction given for raspberry pi.
Thanks,
Atul
@atyadav What’s your project ID? We can take a look and see if we can spot any places to improve your model.
1 Like
Thanks for help. Looks like I modified images now.But I really like to see this use case working accurately…I shall ask once I recreate model.