solin319 commented on a change in pull request #8414: Fix the strides in Gluon
version of ResNet v1
URL: https://github.com/apache/incubator-mxnet/pull/8414#discussion_r152446834
##########
File path: python/mxnet/gluon/model_zoo/vision/resnet.py
##########
@@ -102,10 +102,10 @@ class BottleneckV1(HybridBlock):
def __init__(self, channels, stride, downsample=False, in_channels=0,
**kwargs):
super(BottleneckV1, self).__init__(**kwargs)
self.body = nn.HybridSequential(prefix='')
- self.body.add(nn.Conv2D(channels//4, kernel_size=1, strides=1))
+ self.body.add(nn.Conv2D(channels//4, kernel_size=1, strides=stride))
Review comment:
1. How about add 'use_bias=False' in Conv2D? This will make resnet faster.
The effect of bias in convolution is same as beta in the next batchnorm.
2. Does (2,2) stride set in 1*1 conv or 3*3 conv will effect the final
accuracy of resnet? In 1*1 conv, (2,2) stride will loss some pixels in feature
map.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
