Users should be able to edit autoencoder proposals before annotating them as training data for the FCN. Sometimes the autoencoder only detects part of a bigger object. This can be easily detected by a human observer. The user then should have a convenient way to adjust the proposal to encompass the whole object.
mzur
commented
6 years ago
Users should be able to edit autoencoder proposals before annotating them as training data for the FCN. Sometimes the autoencoder only detects part of a bigger object. This can be easily detected by a human observer. The user then should have a convenient way to adjust the proposal to encompass the whole object.