EricZgw
EricZgw
Have you solved this problem yet? @dianxin556 @MathiasGruber
NotFoundError (see above for traceback): Key feature_fusion/Conv_46/BatchNorm/gamma/ExponentialMovingAverage not found in checkpoint I have the save problem,my tensorflow version is 1.4.1.
it's normal,keep training.
head、body是根据face衍生出来的，论文中叫PyramidAnchors。 我的理解是，低层时负样本（高层时正样本）情况复杂，因此max-in-out留给负样本3个通道（论文中实验得出）将其细分，选择其中最接近正样本的与正样本进行分类，提高分类能力。 @zhangtao22
@samehraban I forgot to associate some parameters with batchsize, and I will modify it later. You can change the code [here](https://github.com/EricZgw/PyramidBox/blob/03f67d8422f2f03e5ee580e2f92d4040c24bbdb4/nets/ssd.py#L728) first. 25600->25600 * 16( your batchsize ) , 6400->6400...
@samehraban The image size in the original paper is 640*640. The pretrained vgg16 checkpoint may not be very useful, I just think it is multi-scale training in a sense. You...
@zh199110 You can try to make [learning_rate_decay_factor](https://github.com/EricZgw/PyramidBox/blob/03f67d8422f2f03e5ee580e2f92d4040c24bbdb4/train_model.py#L61) smaller or change [learning_rate_decay_type](https://github.com/EricZgw/PyramidBox/blob/03f67d8422f2f03e5ee580e2f92d4040c24bbdb4/train_model.py#L401) to 'fixed' and adjust learning rate every few epoches.
I also encountered the same problem. Did you solve it?
Is this a useless project?