Duo Li

@szagoruyko @EderSantana Hi, your sharing code is appreciated, but would you please specify your strategy of decaying the two multipliers α and β during training process? Thanks in advance.

@xiaomengyc Hi, the whole framework adopts modified Inception v3 as the backbone, but why the weights of pre-trained VGG-16 is restored in the training command line [here](https://github.com/xiaomengyc/SPG/blob/master/scripts/train_imagenet_full_v5.sh#L15)? I suppose the...

@xiaomengyc Hi, thanks for releasing the wonderful implementation of your work! I noticed from this snippet since https://github.com/xiaomengyc/SPG/blob/master/exper/my_optim.py#L12, you try to multiply the lr for the last conv layer before...

Resuming from checkpoints for classification scripts

3

Hi, thanks for the great code! I wonder do you have plans to support resuming from checkpoints for classification? As we all know, in terms of training ImageNet, the training...

Default box size of MobileNet-SSD

1

@ruinmessi Hi, thanks for your public released RFBNet-related code! Could you please tell me where you got the settings of default box sizes for SSD or RFBNet with MobileNet as...

Add two papers related to network architecture engineering with source code

About multi-gpu training

5

Thanks for your awesome work! Is there any idea how multi-gpu training is supported? Because you know training ResNet-101 on ImageNet with a single GPU is unacceptably slow.

re-implement Stand-Alone Self-Attention model

2

Hi, @csrhddlam As we discussed before, I am trying to re-implement the baseline "Conv-stem+Attention" in [Stand-Alone Self-Attention in Vision Models](https://arxiv.org/abs/1906.05909), which is referred in your paper. Could you please help...

Update attention.py

3

fix the logical operation of `attn_mask`