How to generate "Fusion data"

[UndecidedBoy]

Hello,

Great work and GitHub repositories!
I was curious regarding the Fusion data generation, could you clarify the arguments (i.e. --separation,--range,..) you used for generating the data with generate_fusion.py?

MiVOS/generate_fusion.py

Lines 24 to 36 in b1992e6

 Arguments loading 
 """ 
 parser = ArgumentParser() 
 parser.add_argument('--model', default='saves/propagation_model.pth') 
 parser.add_argument('--davis_root', default='../DAVIS/2017') 
 parser.add_argument('--bl_root', default='../BL30K') 
 parser.add_argument('--dataset', help='DAVIS/BL') 
 parser.add_argument('--output') 
 parser.add_argument('--separation', default=None, type=int) 
 parser.add_argument('--range', default=None, type=int) 
 parser.add_argument('--mem_freq', default=None, type=int) 
 parser.add_argument('--start', default=None, type=int) 
 parser.add_argument('--end', default=None, type=int) 

Also, it seems that a link is missing for downloading the pre-generated fusion data in the README.md

Best!

[Rex Cheng]

Thanks for reminding me. I have fixed the links.

DAVIS train set (range does not matter):
Set 1 - separation: 5, memory freq: 10
Set 2 - separation: 5, memory freq: 20
Set 3 - separation: 5, memory freq: 30

BL30K:
Set 1 - separation: 30, memory freq: 20, range: 90, start 0, end 5000
Set 2 - separation: 30, memory freq: 20, range: 90, start 5000, end 10000

I generated a few sets for each dataset to increase diversity. These parameters can also be inferred from the names of the folders.

[UndecidedBoy]

Thanks for the quick reply!