Since I didn’t find any regulation about the restriction on using the external data,
I wonder if it is fine to use weights pre-trained on other publicly available datasets
as a initialization. Also, when training a model for a task, can I use annotation and
data for other tasks (in Ego4D dataset)?
(I’m curious about the second question since I find that in the validation report of
the first place in TTM@ECCV2022 they use models for LAM and ASD as a part of their
Yes use of external data (pre-trained weights etc.) is allowed. Also it is OK to use annotation data from other tasks.
However the ultimate decision on challenge winner will be decided by the hosts after evaluating the validation report thoroughly. This is to make sure that the additional data usage is within the challenge rules. For e.g in short term anticipation any method should not be looking at the future frames while making predictions.