benchmark flag for " Short-term object interaction anticipation " about ego4d HOT 2 CLOSED

STA is part of the FHO benchmark, we don't splice the benchmarks by sta/lta/scod yet.

You can download the annotations with

python -m ego4d.cli.cli --output_directory=<OUTPUT_DIR> --datasets annotations

and you'll find the files 'fho_sta_train.json' and 'fho_sta_val.json' in the annotations/ folder.

the benchmark models with

python -m ego4d.cli.cli --output_directory=<OUTPUT_DIR> --datasets sta_models

and the clips with

python -m ego4d.cli.cli --datasets clips --output_directory="~/ego4d_data" --benchmarks FHO

There's more info on the structure of these annotations in the Forecasting docs.

Good catch on the benchmark flags, seems we missed documenting that. I'll add all the options to the cli README.

from ego4d.

I have checked the code, the it says "The benchmarks to download dataset subsets of: 'EM', 'FHO', 'AV'".
'STA' is not among them

from ego4d.