scenic ViT ImageNet Pretrained Model Problem with ViViT

Hey,

I download the pretrained ImageNet21k ViT b_16 model from the urls mentioned in the configuration files, and replace the path in the config file, but for both scenic and bigvision versions I face some problems.

When I put the path to scenic pretrained version, it gives the following error: raise ValueError('No checkpoint for the pretrained model is found in: ' ValueError: No checkpoint for the pretrained model is found in: /content/ViT_B_16_ImageNet21k

When I put the path to bigvision pretrained version, this happens: File "/usr/local/lib/python3.7/dist-packages/scenic/train_lib/pretrain_utils.py", line 269, in convert_bigvision_to_scenic_checkpoint restored_params = tree['opt']['target'] KeyError: 'opt'

Can you please tell me what should I do?

Thanks in advance.

Dec 17 '21 09:12 pabdzadeh

** ** commented on 17 Dec 2021 Hi, I have the same question. Do you have the solution?

Mar 14 '22 19:03 kar9999

We'll look into this soon and send a fix, but in the meantime, I think replacing restored_params = tree['opt']['target'] with restored_params = tree['target'] would solve the issue.

Mar 15 '22 17:03 MostafaDehghani

We'll look into this soon and send a fix, but in the meantime, I think replacing restored_params = tree['opt']['target'] with restored_params = tree['target'] would solve the issue.

Thanks for your answer. When I choose bigvision version, I download it through https://storage.googleapis.com/vit_models/imagenet21k/ViT-B_16.npz Then this happens restored_params = tree['opt']['target'] KeyError: 'opt' Then I replacing restored_params = tree['opt']['target'] with restored_params = tree['target'] . Another problem comes. restored_params = tree['target'] KeyError: 'target'

Mar 16 '22 13:03 kar9999

We'll look into this soon and send a fix, but in the meantime, I think replacing restored_params = tree['opt']['target'] with restored_params = tree['target'] would solve the issue.

Thanks for your answer. When I choose bigvision version, I download it through https://storage.googleapis.com/vit_models/imagenet21k/ViT-B_16.npz Then this happens restored_params = tree['opt']['target'] KeyError: 'opt' Then I replacing restored_params = tree['opt']['target'] with restored_params = tree['target'] . Another problem comes. restored_params = tree['target'] KeyError: 'target'

Have the same issue, which version of the model should I use? Which one works? tree has these attributes only ['Transformer', 'cls', 'embedding', 'head', 'pre_logits']