Skip to content

Conversation

@astorfi
Copy link
Contributor

@astorfi astorfi commented Oct 25, 2017

The first execution of 3x3 convolution filters on the feature map (classify and predict the offset to the default boxes) is applied on Conv4-3 layer which is the 21th element of the vgg list defined by the code. I think, it has been defined 24 by misatke. The 24th element of the vgg list is the Conv5-1 layer of the vgg architecture.

@ASoleimaniB
Copy link
Contributor

I think while it should be [21, -2], it does not change any important thing. At the place you edited the code, they just define the details of loc_layers and conf_layers, and this details is similar if we use 21 or 24, because both 21 and 24 have the same size.
However, in forward function (def forward(self, x)), they apply loc_layer and cong layer at the right place which is conv4_3, vgg[22] after relu.
All in all, your suggestion is OK

@amdegroot amdegroot merged commit afebb30 into amdegroot:master Feb 15, 2018
@amdegroot
Copy link
Owner

Good catch and valid point. While it will not change the functionality, it makes the function more correct and clearer.

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

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants