Release Note
In this release, we are excited to share the weights of Channel Vision Transformers (ChannelViT) trained on datasets as presented in our paper https://arxiv.org/abs/2309.16108
Pre-trained Models
Supervised
| Dataset | Name | Backbone | Hierarchical Channel Sampling |
|---|---|---|---|
| ImageNet | imagenet_channelvit_small_p16_with_hcs_supervised | ChannelViT-S/16 | Yes |
| CP-JUMP (cellpainting) | cpjump_cellpaint_channelvit_small_p8_with_hcs_supervised | ChannelViT-S/8 | Yes |
| CP-JUMP (cellpainting + brightfield) | cpjump_cellpaint_bf_channelvit_small_p8_with_hcs_supervised | ChannelViT-S/8 | Yes |
| Camelyon | camelyon_channelvit_small_p8_with_hcs_supervised | ChannelViT-S/8 | Yes |
| So2Sat (Random Split) | so2sat_channelvit_small_p8_with_hcs_random_split_supervised | ChannelViT-S/8 | Yes |
| So2Sat (Hard Split) | so2sat_channelvit_small_p8_with_hcs_hard_split_supervised | ChannelViT-S/8 | Yes |
DINO
| Dataset | Name | Backbone | Hierarchical Channel Sampling |
|---|---|---|---|
| ImageNet | imagenet_channelvit_small_p16_DINO | ChannelViT-S/16 | No |
Feedback and Contributions
We welcome feedback and contributions from the community. If you encounter any issues or have suggestions for improvements, please submit an issue on the repository.
Thank you for your interest in our work. We look forward to seeing what you will achieve with these model weights!