WXinlong / SOLO

SOLO and SOLOv2 for instance segmentation, ECCV 2020 & NeurIPS 2020.
Other
1.69k stars 307 forks source link

Fine Tuning #167

Closed danielcrane closed 3 years ago

danielcrane commented 3 years ago

Does anyone know if there's any easy way to resume training from for example weights/SOLOv2_R50_3x.pth , however with a different number of classes?

In my case I only have one output class, however of course if I simply set num_classes=2 I get a tensor size mismatch error.

danielcrane commented 3 years ago

Setting load_from= in the config file resolved this.