search

Jyxarthur / flowsam

[ACCV 2024 (Oral)] Official Implementation of "Moving Object Segmentation: All You Need Is SAM (and Flow)" Junyu Xie, Charig Yang, Weidi Xie, Andrew Zisserman
https://www.robots.ox.ac.uk/~vgg/research/flowsam/
Apache License 2.0
267 stars 20 forks source link

Inconsistency in results: additional masks in horsejump-high example #14

Open nnanhuang opened 2 weeks ago

nnanhuang commented 2 weeks ago

Hi, I noticed an inconsistency between the results provided in the README and the project page.

In the example of horsejump-high from the google drive, there are masks not only for the horse and rider but also for several other objects. However, on the project page, the result for the same example only includes masks for the horse and rider.

Could you clarify why there's a difference between these two sets of results?

Thank you!

Jyxarthur commented 2 weeks ago

Hi, thanks for your interest in our work. For Flow-ISAM, we extract object masks with top-5 scores as the results (masks in Google Drive). During evaluation (e.g. on DAVIS16), since there are a lot of other objects (e.g. fences) not labelled by the GT, we Hungarian match the predicted masks with GT masks to calculate IoU scores. Following this, for clearer visualisations, we only illustrate the ones that are Hungarian matched with the GT.