bertinetto / siamese-fc

Arbitrary object tracking at 50-100 FPS with Fully Convolutional Siamese networks.
http://www.robots.ox.ac.uk/~luca/siamese-fc.html
MIT License
619 stars 224 forks source link

What if the input examplar image is less than 127*127? #11

Closed RyanCV closed 7 years ago

RyanCV commented 7 years ago

Hi, the tracker needs to resize the examplar image to 127127 and the next frame to 255255, however, if I want to track small object (e.g. less than 32*32), do you think is it possible? Or any hints on modifying the tracking code? Thanks.

bertinetto commented 7 years ago

Hi, There is no need to modify the code. If the object is smaller than 127x127 it will be simply upsampled to 127x127. In my experience the tracker works particularly well at low resolutions and with small objects, probably because in ImageNet Video (the training set) many objects are very tiny. Hope this helps.

On Mon, Dec 12, 2016 at 10:46 PM, RyanCV notifications@github.com wrote:

Hi, the tracker needs to resize the examplar image to 127127 and the next frame to 255255, however, if I want to track small object (e.g. less than 32*32), do you think is it possible? Or any hints on modifying the tracking code? Thanks.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/bertinetto/siamese-fc/issues/11, or mute the thread https://github.com/notifications/unsubscribe-auth/ADLRualg8LPePRv2SxzhpGnr1xljkU8oks5rHc7TgaJpZM4LLHyp .