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
The essence of FreeAnchor is to select a suitable anchor frame for each object through the free matching of targets and features, and its essence is to choose the appropriate CNN features for each object. Can it be understood that FerrAnchor still needs an artificially designed anchor box, just select the anchor box with a larger score from the anchor boxes created by these people and then use FreeAnchor to process it? In your open source code, I don't see the specific aspect ratio of the anchor box, but in mmdetection, I clearly see anchor_rations=[0.5, 1.0, 2.0].
madexin
changed the title
The essence of FreeAnchor is to select a suitable anchor frame for each object through the free matching of targets and features, and its essence is to choose the appropriate CNN features for each object. Can it be understood that FerrAnchor still needs an artificially designed anchor box, just select the anchor box with a larger score from the anchor boxes created by these people and then use FreeAnchor to process it? In your open source code, I don't see the specific aspect ratio of the anchor box, but in mmdetection, I clearly see anchor_rations=[0.5, 1.0, 2.0].
Does FreeAnchor require artificially designed anchor boxes?Such as anchor_rations=[0.5, 1.0, 2.0]
Jan 11, 2020
No description provided.
The text was updated successfully, but these errors were encountered: