WDHT mAP on 12bit is only 0.3269

I used all the pre-training models and data you provided to test. But on 12 bits, the mAP is only 0.3269. Why is the result so different from that of the paper?

Apr 22 '19 06:04 chenshen03

We will look into this and get back to you. Will be busy this whole week. Please expect a delayed response.

Apr 22 '19 17:04 Vijetha1

@Vijetha1 Thank you for your help. I've been waiting for your reply. Have you found the reason for the problem?

May 06 '19 09:05 chenshen03

Hi Chen, trying to replicate this issue. Will update you asap.

May 06 '19 19:05 Vijetha1

Hi Chen, I was not able to replicate the issue nor the set-up due to the lack of support for Theano. Instead, I recreated the code with Tensorflow backend and it is working as expected. Please verify now.

May 29 '19 02:05 Vijetha1

Thank you very much for your patient answer. I'll try again.

May 29 '19 02:05 chenshen03

Hi, were you able to verify this?

Aug 30 '19 20:08 Vijetha1

HI ，I cannot find the dataset and pretrain weight ，can you help me?

Sep 21 '19 07:09 ximi2018

@ximi2018 Can you download the dataset and weight? I cannot download them. Could you tell me how to deal with it?

Nov 28 '19 08:11 mengliu1991

@mengliu1991 sorry,I cannot get them

Nov 28 '19 08:11 ximi2018

@mengliu1991 @ximi2018 Maybe you need a VPN. The link provided by the author is still available for download.

Nov 28 '19 15:11 chenshen03

@chenshen03 Hi, Have you test the L3 loss+L4 loss case, i.e., using the similarity matrix to train the model? I change the model into the binary tag-vector model, but I cannot obtain the result reported in the paper.

Dec 17 '19 00:12 mengliu1991