ChaoningZhang
commented
1 year ago My guess is that vit_h was trained with half precision, while vit_b and _vit_l are not, which can be the reason. Since the original paper did not provide such details, this is just a guess. Note that that our tiny-vit was also trained by distillation of full precision. If you find out the real reasons for why this happened, please kindly let us know
just like vit_h, MobileSAM doesn't work well when using half precision, but vit_b and _vit_l can work normally, have you ever thought about the reason of this phenomenon?It would be thankful if you can reply