Implementation of paper - YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors
GNU General Public License v3.0
13.34k
stars
4.2k
forks
source link
Re-parameterization have a bug #450
Closed
JasonDu1993 closed 2 years ago
https://github.com/WongKinYiu/yolov7/blob/main/tools/reparameterization.ipynb, I set conv m get the output through [(x+ia) m_weight + m_bias] im, so the new_weight of conv m is m_weight im.implicit, the new_bias is (ia m_weight im+ m_bias im),
but the file ipynb set new_weight is m_weight im.implicit, the new_bias is (im m_weight ia + m_bias ) im = (ia m_weight im im + m_bias im) ,
because the weight is changed after the for loop, so I think the result is wrong
the state_dict['model.105.m.0.weight'] is old weight ,sorry I hadn't noticed it before