-> keep_inds = torch.cat([fg_inds, bg_inds], 0) (Pdb)

[qujiangang]

-> keep_inds = torch.cat([fg_inds, bg_inds], 0)
(Pdb)

[ChrisHJC]

Do you want to insert some conv layers？Maybe you forget give weight at net/network.py line478.

[zhilaAI]

how to fix it?

[happybored]

Either voc2007 or my own data has the same question!!

[ChrisHJC]

I got this question because I want to insert another convolution layers，so I fix it by giving these layers weights.

[happybored]

But I didn't revise anything_(:з」∠)_

…

---Original--- From: "ChrisHJC"<[email protected]&gt; Date: Mon, Nov 11, 2019 21:20 PM To: "ruotianluo/pytorch-faster-rcnn"<[email protected]&gt;; Cc: "happybored"<[email protected]&gt;;"Comment"<[email protected]&gt;; Subject: Re: [ruotianluo/pytorch-faster-rcnn] -&gt; keep_inds = torch.cat([fg_inds, bg_inds], 0) (Pdb) (#155) I got this question because I want to insert another convolution layers，so I fix it by giving these layers weights. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.

[zhilaAI]

where we should Add these lines? I added them to network.py but does not work again.
in faster rcnn we do not have query and ... and also i checked that lines we have the same parameter for normal_init.

[jc56968]

I have the same question,because i use the res50.pth which is downloaded from other website,I fix it by downloading the res50.pth from google driver the author gived.it may help you.

[zhilaAI]

You mean we should use specific type of res net for the model? If I want to test the model on VGG16, it doesn't work.