Closed devraj89 closed 3 years ago
Hi Devraj89, you can try the model in any way you like! For example, you can just load the model from glasses and use a code similar to "train a classifier" from pytorch.
I do not train the model from scratch, since I will need a enormous amount of GPUs, I just copy the weights from other models and then I evaluate the resulting model on ImageNet :)
Hi
The library is very cool and informative.
I am wondering are you planning to share the training script?
I had a look at your tutorial of VIT in https://github.com/FrancescoSaverioZuppichini/ViT and its wonderful.
I am wondering did you train all these models from scratch or simply transferred the weights from the pre-trained available models. I have heard training a VIT is quite challenging. Alternatively, I guess I could use the transfer_learning notebook as a boiler plate code to train it ?
Thanks Devraj