Get the value of a resistor from a picture.
pic2ohms works in three stages: first a neural network scans the input picture and localizes resistors. These resistors are passed to a second neural network that tries to detect the orientation they are in. Finally a group of three neural networks identify the colors of the bands on each resistor (so far it ignores the tolerance ring or resistors with four bands).
Right now the scripts that generate the training data produce images with very little variation, with only a few hundred possible outputs, so of course the NN overtrain and get 100% accuracy. The next step is then to improve the data generation.