Conversation
efficientNet imports
config edit path for eff3 and dropout for eff0
|
|
|
Thanks for your contribution! @jiangyitong Please sign the CLA first. And it seems that your local git email address is not added to your GitHub account, please follow this guide https://help.github.com/articles/why-are-my-commits-linked-to-the-wrong-user/#commits-are-not-linked-to-any-user |
| ] | ||
| norm_cfg = dict(type='SyncBN', requires_grad=True, eps=1e-3, momentum=0.01) | ||
| model = dict( | ||
| pretrained='open-mmlab://efficientnet_b0', |
There was a problem hiding this comment.
Has this weight been uploaded to openmmlab?
There was a problem hiding this comment.
No, we need to convert the weight from facebookresearch to mmcv version
| warmup_ratio=0.1, | ||
| warmup_by_epoch=True) | ||
| fp16 = dict(loss_scale=512.) | ||
| custom_hooks = [dict(type='EMAHook')] |
There was a problem hiding this comment.
It is recommended to use mmdet's LinearMomentumEMAHook
custom_hooks = [dict(type='LinearMomentumEMAHook', skip_buffers=True)]
There was a problem hiding this comment.
The evaluation IOU is 0 when adding 'LinearMomentumEMAHook'
| self.conv1 = build_conv_layer( | ||
| self.conv_cfg, | ||
| in_channels, | ||
| out_channels, | ||
| kernel_size=3, | ||
| stride=2, | ||
| padding=1, | ||
| bias=False) | ||
| self.norm1_name, norm1 = build_norm_layer( | ||
| self.norm_cfg, out_channels, postfix=1) | ||
| self.add_module(self.norm1_name, norm1) | ||
| self.swish = Swish() |
There was a problem hiding this comment.
Can these be replaced with a ConvModule?
efficientnet review2
| @@ -0,0 +1,53 @@ | |||
| import argparse | |||
There was a problem hiding this comment.
move it to tools/model_converters/
fix model converter and efficientnet import
register memoryefficietnswish
pull master
eff0, eff3
…tion into efficientnet
fix flake8 yapf isort
same padding in tensorflow and edit drop path rate
update mmdetection to master
|
Hello, When can mmsegmentation use Efficientnet as the backbone network? |
You should ask it in MMSegmentation. |
|
supported in #7514 |
Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers.
Motivation
Add EfficientNet-b0 + FPN and EfficientNet-b3 + FPN
Modification
add EfficientNet model in the mmdet/models/backbone/efficientnet.py
add two configs for efficientNet-b0 and efficientNet-b3 training
BC-breaking (Optional)
acknowledgement
Thanks for David de la Iglesia Castro's contribution. We are grateful that he helps to implement the base efficientNet model and the corresponding tools: https://github.com/daavoo/mmdetection
Use cases (Optional)
Checklist