After Freezing the Vit backbone how many iterations are supposed to be trained?

[iacopo97]

Hi, thank you for your fantastic work. I am working with Mask2former framework, and I am training from scratch according to the Mapillary case. The only difference with your framework is that I have freezed the ViT transformer backbone, do you have any suggestion about the number of iterations that I am supposed to use during the training procedure? Should they be 80K or I can reduce this number?

[bysowhat]

Hi, thank you for your fantastic work. I am working with Mask2former framework, and I am training from scratch according to the Mapillary case. The only difference with your framework is that I have freezed the ViT transformer backbone, do you have any suggestion about the number of iterations that I am supposed to use during the training procedure? Should they be 80K or I can reduce this number?

Hi, I want to freeze the vit backbone too. Have you any experiment results? Will it cause low performance?