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
Hey, I noticed when converting YoloV4-tiny that in code at lines 101-102, anchors with indices 1,2,3 are passed to the detection layer instead of anchors 0,1,2. I understand that in default pretrained YoloV4 tiny, the mask is wrong in the CFG and set to 1,2,3, which impacts the detection of small objects. This is fixed with CFG for custom objects.
So, a good way might be to provide an option to just use right indices 0:3 and 3:6, and then fix this with anchors flag as input.
The text was updated successfully, but these errors were encountered:
Hey, I noticed when converting YoloV4-tiny that in code at lines 101-102, anchors with indices 1,2,3 are passed to the detection layer instead of anchors 0,1,2. I understand that in default pretrained YoloV4 tiny, the mask is wrong in the CFG and set to 1,2,3, which impacts the detection of small objects. This is fixed with CFG for custom objects.
So, a good way might be to provide an option to just use right indices 0:3 and 3:6, and then fix this with anchors flag as input.
The text was updated successfully, but these errors were encountered: