You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
How to add background images? ... just drop them in train and val directories
The wiki page and some posts by @glenn-jocher showing that indeed it improves the results.
However, when I look in the code I am not able to find why or rather how it helps!
So far based on my reading of the code, in the build_targets, the samples with no associated labels will be dropped. If this understanding is correct then this further implies that there will be no loss contributed to the total loss (of a batch) by the background samples.
Let's say the batch size is 64. In this batch, 2 samples are so called background images.
Does this mean that the effective batch size is 62? How does that help in getting better results?
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
-
Hi,
I understand following:
However, when I look in the code I am not able to find why or rather how it helps!
So far based on my reading of the code, in the build_targets, the samples with no associated labels will be dropped. If this understanding is correct then this further implies that there will be no loss contributed to the total loss (of a batch) by the background samples.
Let's say the batch size is 64. In this batch, 2 samples are so called background images.
Does this mean that the effective batch size is 62? How does that help in getting better results?
Please guide.
Regards & thanks
Kapil
Beta Was this translation helpful? Give feedback.
All reactions