chaoma99 / dslt Goto Github PK

Hello, thanks for your work.
I download the code and when I run OTB sequence Basketball, the boundding box shinks all the time, I get a point finally, and the final result is bad.Can you explain it for a little bit? thank!

would the training code be released? I did not find any details about offline training in this paper, e.g. the offline training dataset?

Hi, thanks for your work DSLT in ECCV18. After reading the code, I have found several parts of the code are not corresponding with that in the published paper. Can you explain these for more details?

• Backbone net. The backbone net of the tracker is based on VGG, but two maxpooling layers are removed. Thus the conv3_3 conv4_3 conv5_5 share the same featuremap resolution and no more deconv is needed. Will this help locate object more precisely with more details?

• Another block (seems like channel attention module) is added. This block (with global pooling and channel reduction-increase) is added on conv4_3 and conv5_3 before they are used for feature fusion and response generation. I think it is used for channel attention.

• Loss function. In the shrinkage loss definition. label Y is multiplied with a constant 1.6 "label_exp = exp(1.6*label);" in the code, which is mot mentioned in the paper.

Can you spare minutes to explain the modifications in the code for me? Thanks very much ! @carrierlxk @chaoma99

Hello, thanks for your work.
How much your FPS? Thank! (:

Hi, I try to run the tracker with vot-toolkit at the stack of vot2016 on matlab with an 1080ti GPU, when the tracker process the image 17 in the 'bag' sequence, there is an issue as below,

@@TRAX:state 
"334.8400,114.7000,380.6600,134.4600,332.7500,245.5600,286.9300,225.8100" 
@@TRAX:frame "file:///home/vot2016/bag/00000018.jpg" 
F0414 11:34:51.577589 19789 syncedmem.cpp:56] Check failed: error == cudaSuccess (2 vs. 
0)  out of memory

And when I run the demo code run_DSLT.m with the sequence in OTB_dataset, the tracker can produce the outcome normaly. It is just so strange!!

Hi, good work as always.
When do you plan to upload the code?
Looking forward to understand your work detailedly by code~
Thank u very much.

Thanks for your great job!! Could you please provide the results on dataset UAV123?

I want to compare DSLT tracker vs my tracker, unfortunately I cannot access BaiduPan, because of restriction of download for non-Chinese people, can you upload your result on Github, dropbox or anything that is accessible for all researchers.
Thanks