Masked Training best practices? #67
-
|
How should I be captioning images when using masked training? For example, say I was training images of wristwatches being worn and wanted to get consistent accurate results on different types of wrist watches- if the watch is in the whitespace of the mask I am assuming that is what the model is mostly focusing on training (depending on what settings you set in the masked training options), so should the captions reflect what is in the white space or the whole picture? Or is there a way to caption for the whitespace and blackspace separately? Masked training is pretty new to me, but the tool that was added to auto generate and manually edit the mask has been pretty nice to use. |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 1 reply
-
|
The captions should reflect only to what is in the mask and have been targeted. Visualise that everything outside of the mask won't be seen that's the point of the masking process. |
Beta Was this translation helpful? Give feedback.
-
|
black or white area? |
Beta Was this translation helpful? Give feedback.
The captions should reflect only to what is in the mask and have been targeted. Visualise that everything outside of the mask won't be seen that's the point of the masking process.