Padding issue of SiamFc

Zhongdao / UniTrack

[NeurIPS'21] Unified tracking framework with a single appearance model. It supports Single Object Tracking (SOT), Video Object Segmentation (VOS), Multi-Object Tracking (MOT), Multi-Object Tracking and Segmentation (MOTS), Pose Tracking, Video Instance Segmentation (VIS), and class-agnostic MOT (e.g. TAO dataset).

MIT License

334 stars 35 forks source link

Padding issue of SiamFc #32

Open Morefre opened 2 years ago

Morefre commented 2 years ago

Hi! When evaluating the appearance models on SOT using SiamFC tracker, how does padding operation, which breaks SiamFC's fullly-convolutional property, affect the result? Thanks in advance!

Zhongdao commented 2 years ago

Hi @Morefre , Good question! Considering that all the off-the-shelf SSL models we tested as appearance models are trained with zero padding, we did not evaluate how much it improves by removing zero paddings. But I guess it definitely will improve results in propagation-type tasks like SOT and VOS.