Skip to content

Recurrence problem of Youtube-vis2019 #31

Description

@kamdkslansgka

Hello author, thank you for your excellent work.
I would like to ask, I can't seem to reproduce the accuracy in the paper in my experiment. The accuracy of Youtube-2019 in the paper is 55.5 (Resnet50), but I reproduced it and it is only 51. The following is the log of my training. Is there any problem?

log.txt

The training script is as follows:

python /data/Share/Student/2024/ChenYuran/experiment/DVIS_Plus-main/DVIS_Plus/train_net_video.py
--num-gpus 8
--config-file /data/Share/Student/2024/ChenYuran/experiment/DVIS_Plus-main/DVIS_Plus/configs/dvis_Plus/ytvis19/DVIS_Plus_Online_R50_8wIter.yaml
--resume
MODEL.WEIGHTS /data/Share/Student/2024/ChenYuran/experiment/DVIS_Plus-main/DVIS_Plus/ctvis_r50_ytvis19.pth
SOLVER.IMS_PER_BATCH 8

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions