ConvNext B for reconstruct images

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

1.41k stars 82 forks source link

ConvNext B for reconstruct images #75

Closed M-SunRise closed 4 months ago

M-SunRise commented 6 months ago

Can you provide the weights of ConvNext-B for me?

keyu-tian commented 6 months ago

Hi @M-SunRise, you can click and expand the "Pretrained weights (with SparK's UNet-style decoder; can be used to reconstruct images)" in README for downloading them.

M-SunRise commented 6 months ago

I only find ConvNext-S and L, not find B.

keyu-tian commented 6 months ago

@M-SunRise oh I remember now. I didn't save the convnext-B's checkpoint with decoder. So currently only S and L are available for reconstruction.