megvii-research / MOTRv2

[CVPR2023] MOTRv2: Bootstrapping End-to-End Multi-Object Tracking by Pretrained Object Detectors
Other
348 stars 44 forks source link

num_classes设置 #14

Closed hahapt closed 1 year ago

hahapt commented 1 year ago

加载pretrain模型r50_deformable_detr_plus_iterative_bbox_refinement-checkpoint.pth时会出现报错: No param refine_embed.weight.If you see this, your model does not fully load the pre-trained weight. Please make sure you set the correct --num_classes for your own dataset.

请问在哪里更改num_classes设置?

zyayoung commented 1 year ago

不需要修改,忽略这条报错即可。