Missing video IDs when downloading full_scale videos

[sanketsans]

Hi,
I am downloading the full_scale videos from the given cli :
python -m ego4d.cli.cli --output_directory=<path> --dataset full_scale--benchmarks FHO --version v2

However, when I run the code from fho_sta_train.json to dump the frames to lmdb files from here. It give error for missing videos.
eg. aad332b0-a437-4bd9-9147-f580bb58c3f5, 64747fa7-c7b9-46fc-ac9d-ea9a514d91df -( missing video uid )

There are many such video uid found. I also tried to separately download the full_scale for v1 and v2. These videos are not found in either of them. I am using 'fho_sta_train.json' extracted from annotation download for v2.

Can you please check if something is wrong ?

Thanks. :)

[sanketsans]

Just noticed that the IDs are present in manifest.csv but the cli command is not downloading them, since they are marked for other splits ([EM|MQ|VQ]), but not for FHO. However, these videos are present in the fho_sta_train.json files. These IDs are also not present in object_detection.json file which is downloaded for FHO benchmark.
Is there a way to either eliminate these IDs / include them.
Can we modify it ?

[antoninofurnari]

Hi, thank you for opening the issue and suggesting a solution, much appreciated!
Maybe @ebyrne can help or tag who can?

Antonino

[Tushar-N]

Relevant issue and workaround here