New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fix names of extract-layers of resnet101 #8998

Merged
merged 1 commit into from Dec 8, 2017

Conversation

Projects
None yet
2 participants
@RobinDong
Contributor

RobinDong commented Dec 8, 2017

Description

After running resnet50 and resnet101 in example/ssd, the size of .params file for these two models are:

Model Size of .params file
resnet50 119MB
resnet101 69MB

Commands:

python train.py --network resnet50
python train.py --network resnet101

It shows the number of parameters for resnet101 is smaller than resnet50, which seems very weird.

Changes

  • The names of extract-layers for resnet101 in symbol/symbol_factory.py

Comments

For resnet50, the numbers of layers for different stages are [3, 4, 6, 3], so the extract-layers for SSD are 3+4+6-1=12 and 3+4+6+3-1=15,as the code in symbol_factory.py:
from_layers = ['_plus12', '_plus15', '', '', '', '']
The numbers of layers for resnet101 are [3, 4, 23, 3], so the extract-layers for SSD should be 3+4+23-1=29 and 3+4+23+3-1=32.
The right code for resnet101 should be:
from_layers = ['_plus29', '_plus32', '', '', '', '']

@piiswrong piiswrong requested a review from zhreshold Dec 8, 2017

@zhreshold

Thanks for figuring this out.

@zhreshold zhreshold merged commit 075ef68 into apache:master Dec 8, 2017

2 checks passed

continuous-integration/appveyor/pr AppVeyor build succeeded
Details
continuous-integration/jenkins/pr-head This commit looks good
Details

zhreshold added a commit to zhreshold/mxnet that referenced this pull request Dec 14, 2017

rahul003 pushed a commit to rahul003/mxnet that referenced this pull request Jun 4, 2018

zheng-da added a commit to zheng-da/incubator-mxnet that referenced this pull request Jun 28, 2018

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment