Luca Bertinetto

Hi, it is computing the similarity between target and search area. As input, it gets the precomputed embedding of the template (because they are fixed), and the rgb crops of...

We haven't tried it yet. Please feel free to report bugs if they arise.

Hi, I think you can find inspiration from here: https://github.com/bertinetto/staple/blob/master/runTracker_webcam.m Hope it helps!

We used the tiniest variant of ResNet (I think 18), which is actually leaner than VGG architectures. Also, consider that the tracking speed is affected only by the inference time....

Hi, unfortunately we are not able to support newer versions of Matconvnet. Does this error arise from the code as it is or did you modify something?

Hi, not that I am aware of.

Hi, Yes, your interpretation is correct. On Fri, Feb 15, 2019 at 8:56 AM CoinCheung wrote: > Hi, > > Thanks for providing this helpful codebase. I am trying to...

Modify the network with dilate field set to 0 should work. Anyway, I highly recommend to use the SiamFC v2 code of CFNet (referred in paper as baseline-conv5). https://github.com/bertinetto/cfnet

Hi, What do you mean by that? Are you re-computing the exemplar at every frame?

Hi, Apologies for the late answer: I was at CVPR + holidays. 1) Yep we do upsample the score map during tracking. 2) Yes we want to limit the stride...