search

keyu-tian / SparK

[ICLR'23 Spotlight🔥] The first successful BERT/MAE-style pretraining on any convolutional network; Pytorch impl. of "Designing BERT for Convolutional Networks: Sparse and Hierarchical Masked Modeling"
https://arxiv.org/abs/2301.03580
MIT License
1.42k stars 82 forks source link

Can you provide more model pth with decoder? #29

Closed WwwwYz666 closed 1 year ago

WwwwYz666 commented 1 year ago

What an outstanding and admirable work! Due to performance limitations, this experiment cannot be trained on a PC. But I want to try to see the effect of image restoration using different models. I would like to ask if you can provide more pretrain model pth files including decoder.

Just like the pth model in https://github.com/keyu-tian/SparK/blob/main/pretrain/viz_reconstruction.ipynb step 3 .

keyu-tian commented 1 year ago

Thank you! For other pths you can find them in Pretrained weights (with SparK's UNet-style decoder; can be used to reconstruct images) section in https://github.com/keyu-tian/SparK. We also updated the https://github.com/keyu-tian/SparK/blob/main/pretrain/viz_reconstruction.ipynb, so you can use different models like ResNet50 for reconstruction.

WwwwYz666 commented 1 year ago

I sincerely appreciate your help!