D2Det icon indicating copy to clipboard operation
D2Det copied to clipboard

Published 20 hours ago • verified publisher icon JialeCao001

Discriminative ROI Pooling

Open Kevin43614 opened this issue 4 years ago • 5 comments

Hello I want to ask one question about your paper. You say you use a pooling size of 7 × 7 (where k = 7) for classification, so "light-weight offset prediction only requires a k/2 ×k/2 sized RoIAlign" which means pass 3.5*3.5's feature map through fully connected layers ?

Kevin43614 avatar Nov 15 '20 14:11 Kevin43614

@Kevin43614 Thanks for interest. Yeah. I remeber that we use 3x3 for offset prediciton.

JialeCao001 avatar Nov 17 '20 05:11 JialeCao001

@JialeCao001 If the input size used to offset prediction is 3x3 , and through fully connected layers , how to do RoIAlign and generate a 2k2k(1414) size feature map ?

Kevin43614 avatar Nov 17 '20 10:11 Kevin43614

@Kevin43614 After fc layers, we reshape the vector to feature map and upsample the feature map.

JialeCao001 avatar Nov 17 '20 12:11 JialeCao001

@JialeCao001
I am also confused about this part. Can you provide more details about operations from three fc layers to generate (2k x 2k) resolution feature map?

z0978916348 avatar Nov 17 '20 15:11 z0978916348

@z0978916348 Please refer the code. https://github.com/JialeCao001/D2Det/blob/a76781ab624a1304f9c15679852a73b4b6770950/mmdet/ops/dcn/deform_pool.py#L199

JialeCao001 avatar Nov 18 '20 03:11 JialeCao001