ltl

How could I use optical flow to estimate proposal motion?

1

Hi, sorry to bother you again. I'm commited to solving the existing problems in OWT proposed by your paper. As mentioned in your paper, the baseline uses both optical flow...

I use the config "configs/gfocal/gfocal_r50_fpn_1x.py" with the use_sigmoid equals False, just like the pic shown as follows:  then the bg_classes_id got 81, this means all the proposals will be...

Which config is used by OWTB in the final experiment?

1

Hello, By disturb you again. May I ask if the final experiment result of OWTB comes from the config "Detectron2-OWT/Configs/New_BASELINES/MASK_RCNN_R_101_FPN_400EP_LSJ_OWT.PY" ?

Where is this "resnet" imported from?

3

I tried replacing it with torchvision.models.resnet, but that doesn't seem to satisfy subsequent code, e.g. "resnet imported from torchvision" does not have a "modify" method.

How to run the segmentation demo?

3

There seems to be a config file"models/upernet_swin.py" missing in the folder "Track1/segmentation/configs/_base_".