Closed weihua93 closed 5 years ago
Hi @weihua93 How many epochs you conduct when the accuracy is 0.869? Thanks
Hi @weihua93 How many epochs you conduct when the accuracy is 0.869? Thanks
183
Thanks
hi, I have some problems on the pretrained model of ResNet50 on ImageNet, which pretrained model did you use, the 224224 one or the 448448 one?
I used 448
I used 448
Thank you! While I can't find the pretrained model with size of 448, Can you send me one? my email is hywu@mail.bnu.edu.cn
The provided one is 448
The provided one is 448 you mean here provided the one of 448, I just can't open the link
I have sent your email. Please check it.
Let me know your result from scratch if you train model
I have sent your email. Please check it.
Let me know your result from scratch if you train model Sorry,I still can't open the link you send me. So sorry to touble you.
+1, 试了多种batch_size,多卡组合,准确率最高只能到0.868
I got the best accuracy of 87.3% (K=4) in epoch 127 following default setting. Although the accuracy fluctuates between 86.8% and 87.3% from epoch 127 to epoch 156 (I stopped in epoch 156).
@MichaelLiang12 Did you use the defaulted parameters to get 87.3%? or made some change on the config ?
Just as I said, I followed default setting with no changes.
batchsize=16,两块GPU训练,学习率0.001,达不到论文中的0.875的准确率。请问,你们达到的是怎么设置的参数?训练了多少epoch?