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
In the paper, you said the threshold of 0.9 for VOC and 0.5 for Context dataset
but in the code, the threshold is
test_cfg = dict(bg_thresh=.95, mode='slide', stride=(224, 224), crop_size=(448, 448)) for VOC
and
test_cfg = dict(bg_thresh=.35, mode='slide', stride=(224, 224), crop_size=(448, 448)) for Context
Why this is inconsistent? if it is because my misunderstanding, could you please point out what is the correct why to check the background threshold?
The text was updated successfully, but these errors were encountered:
Hi @junchen14
The released models are re-produced version after submission. So some hyperparameters are different. That's also why the accuracy is higher than reported.
In the paper, you said the threshold of 0.9 for VOC and 0.5 for Context dataset
but in the code, the threshold is
test_cfg = dict(bg_thresh=.95, mode='slide', stride=(224, 224), crop_size=(448, 448)) for VOC
and
test_cfg = dict(bg_thresh=.35, mode='slide', stride=(224, 224), crop_size=(448, 448)) for Context
Why this is inconsistent? if it is because my misunderstanding, could you please point out what is the correct why to check the background threshold?
The text was updated successfully, but these errors were encountered: