Content of this repo:
- Data set of 1'840 images which contain one or more signals
- OpenLabeling Software to label the images OpenLabeling
- test statistic files per 5000 iterations
- notebook which creates train/test split
- notebook which creates graphs bases on test statistic
pip install -r requirements.txtTrue Positive = Signals which got predicted right. Example: we have a 6, we predict a 6 which is true
False Positive = Signals which got predicted wrong. Example: we have a 6, but we predict a 4 which is wrong
False Negative = Signals which weren't detected at all
Most important graph!
The quite large amount of false positives comes from double predictions on single signals:
Example: We have a signal info-6, we predict ...
| probability | class |
|---|---|
| 80 % | info-6 |
| 50 % | info-9 |
In this example, it is clear that 6 is the correct answer. The second answer (50% info-9) has a high enough probability, so it adds to the false positives.
info signals are the higher ones
stop signals are the lower ones
