Finetuning instead of training

SIAnalytics / RHINO

Official implementation of the paper "Hausdorff Distance Matching with Adaptive Query Denoising for Rotated Detection Transformer"

Other

25 stars 0 forks source link

Finetuning instead of training #6

Closed Fritskee closed 2 months ago

Fritskee commented 2 months ago

Hi @nijkah

I can understand how we can freeze the different stages of the resnet50 backbone, but when using the swin v2 transformer as a backbone, there is no freeze option.

When I go to look into mmrotate/models/backbones I also don't see a file for swin, but only for resnet.

Could you help me clarify this?