search

dvl-tum / motsynth-baselines

Baselines and setup instructions for the MOTSynth dataset (ICCV 2021)
MIT License
29 stars 11 forks source link

About the challenge #11

Open SangbumChoi opened 2 years ago

SangbumChoi commented 2 years ago

Hi author of MOTSYNTH Can we use real data (e.ge. MOT17) with un-labeled and use unsupervised learning?

guillembraso commented 2 years ago

Hi,

Sorry for the late reply. We've had an internal discussion regarding this, and have decided that we will allow it!

Guillem

voldemortX commented 2 years ago

@guillembraso Does that mean UDA (Unsupervised Domain Adaptation) can be applied with MOT17-train as target domain? Also, does this unsupervised learning still exclude MOT17-test?

guillembraso commented 2 years ago

Hi @voldemortX,

Does that mean UDA (Unsupervised Domain Adaptation) can be applied with MOT17-train as target domain?

Yes, this is allowed!

Also, does this unsupervised learning still exclude MOT17-test?

Yes, MOT17-test cannot be used, even within unsupervised frameworks. Any solution using MOT17-test data in any form during training will not be considered for the challenges.

Best,

Guillem

voldemortX commented 2 years ago

@guillembraso Thanks. What about other sources of real data. I mean is mot17-train still the only allowed data even in a unsupervised manner?

guillembraso commented 2 years ago

Hi @voldemortX,

What about other sources of real data. I mean is mot17-train still the only allowed data even in a unsupervised manner?

Yes, no other sources of real data beyond MOT17 can be used. We have decided to set this constraint in order to keep synthetic as the main source of training data and have a fair comparison among competitors.

Best,

Guillem

chenrxi commented 2 years ago

hi, the link to motsynth dataset is broken, is there new link?Also, the domain gap between motsynth(MOTA:54) and mot17(mot17 half train only, MOTA:60) is a little small, is it necessary for domain adaption. @guillembraso