Closed atonyo11 closed 7 months ago
We always set the temporal kernel size as 1, and thus it's still a 2D CNN.
---Original--- From: @.> Date: Mon, Nov 13, 2023 21:14 PM To: @.>; Cc: @.***>; Subject: [hulianyuyy/CorrNet] 2D-CNN or 3D-CNN? (Issue #15)
Sorry if I miss understand. As show in the picture, feature extractor is 2D CNN, but in the code, resnet based is build from 3D CNN, eg. conv3x3, BatchNorm3d etc. So why it is feature extractor is 2D CNN? Thank you in advance.
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID: @.***>
temporal kernel size as 1
As I understand, when kernel_size=(1,7,7) in the line self.conv1 = nn.Conv3d(3, 64, kernel_size=(1,7,7), stride=(1,2,2), padding=(0,3,3), bias=False)
, it functions as a 2D CNN. Is it right?
yes, it's right.
---Original--- From: @.> Date: Tue, Nov 14, 2023 16:52 PM To: @.>; Cc: @.**@.>; Subject: Re: [hulianyuyy/CorrNet] 2D-CNN or 3D-CNN? (Issue #15)
temporal kernel size as 1
As I understand, when kernel_size=(1,7,7) in the line self.conv1 = nn.Conv3d(3, 64, kernel_size=(1,7,7), stride=(1,2,2), padding=(0,3,3), bias=False), it functions as a 2D CNN. Is it right?
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>
I got it. Thank you very much!
I apologize if I misunderstood. As shown in the picture, the feature extractor is depicted as a 2D CNN. However, in the code, a ResNet-based model is built using 3D CNN components, such as conv3x3, BatchNorm3d, etc. Could you please explain why the feature extractors are implemented using 2D CNN? Thank you in advance.