search

pjh5672 / YOLOv1

YOLOv1 implementation using PyTorch
GNU General Public License v3.0
6 stars 1 forks source link

Neck Layer FCs #3

Open joonesmks opened 1 year ago

joonesmks commented 1 year ago

In the original paper (https://arxiv.org/abs/1506.02640) the authors are using two fully connected layers for the "neck" layer, but I can’t find them in this implementation.

Instead the neck in this implementation neck.py is made of convolutions. What is the rationale behind this?

pjh5672 commented 1 year ago

@joonesmks As you know, It has some tweaks from the paper's network design since I got low mAP scores rather than the paper's one. Thus, I make the model extract richer features from 5 conv layers in neck instead of two fully-connected layers.