search

jixing0415 / caffe-mobilenet-v3

Caffe Implementation of MobileNets V3
MIT License
130 stars 94 forks source link

mobilenet v3 trian #3

Open taoxinlily opened 5 years ago

taoxinlily commented 5 years ago

Why did you use relu6 instead of relu after scale layer in blocks?I think the paper used relu.How many batches when you train? Can you release your train.prorotxt and solver.prototxt?I can't reproduce your result.Thank you!

jixing0415 commented 5 years ago

Why did you use relu6 instead of relu after scale layer in blocks?I think the paper used relu.How many batches when you train? Can you release your train.prorotxt and solver.prototxt?I can't reproduce your result.Thank you!

I think you are right. I will modify the use of relu layer. When I was training, I set the batch size to 64, and I used 8 gpus for parallel training. I will release the pre-train model and train.prototxt ASAP.

xjhjinhui commented 5 years ago

Could you please release the train.prototxt and pre-train model?Thanks a lot!

XeChen commented 4 years ago

Could you please release the train.prototxt and pre-train model?Thanks a lot!

1274622976 commented 3 years ago

Why did you use relu6 instead of relu after scale layer in blocks?I think the paper used relu.How many batches when you train? Can you release your train.prorotxt and solver.prototxt?I can't reproduce your result.Thank you!

I think you are right. I will modify the use of relu layer. When I was training, I set the batch size to 64, and I used 8 gpus for parallel training. I will release the pre-train model and train.prototxt ASAP. Hello, can you send the trained network model （mobilenetv3）to my email (1274622976@qq.com)