evaluating offline model in robotap raise oom error

[Binyr]

Hi, thanks for your great work!

When I evaluate offline model on RoboTap dataset, it raises OOM error. It seems some video of RoboTap is too long (e.g., T>2k).

How can I evaluate offline model on RoboTap dataset?

My evaluation command line is :
python ./cotracker/evaluation/evaluate.py --config-name "$cfg"
offline_model=True window_len=60
exp_dir="${EXP_BASE}/${cfg}"
dataset_root="${DATASET_ROOT}"
checkpoint="${CKPT}"
num_uniformly_sampled_pts=1000
done

Thank you,
Yanrui