dinov2
dinov2 copied to clipboard
PyTorch code and models for the DINOv2 self-supervised learning method.
When training without providing the `mixed_precision` argument to FSDP, there is an error related to dtype mismatch in `dinov2/layers/block.py`. Is this expected? Full stacktrace: ```txt File "/.venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in...
I'm still not entirely sure how ViTs learn various features as I've always intuitively understood convnets better. How can I get feature maps that include the color? For my use...
Do the authors have any experiments with training or distilling other backbones, e.g. ResNets?
pip install command of notebooks makes error because REPO_URL = "https://github.com/facebookresearch/dinov2[extras]" is not a valid editable requirement. Thus, changed to REPO_URL = "git+https://github.com/facebookresearch/dinov2.git#egg=dinov2"
With regards to this sentence: > In Table 5, we show that the Top-1 accuracy on the validation set of ImageNet-1k improves by more than +2% when the backbone is...
It would be very much appreciated if the code for patch matching (mentioned on page 18 in the paper) could be shared using DINO V2.
I download the 'dinov2_vitl14_pretrain.pth' locally and use the 'vit_large' function to build model. I run the fellowing code. But the weight and the model are not compatible ```python from dinov2.models.vision_transformer...
Sorted imports by 1.`isort`; 2.`black` (order is important)
i trained dinov2_vitl14 with custome data that add class on imagenet-1k,and use eval/teacher_checkpoint.pth,and run DINOV2 PCA demo on huggingface,got below:Traceback (most recent call last): File "/home/data-lv/luolu/PycharmProjects/dinov2-pca-master/venv/lib/python3.10/site-packages/gradio/routes.py", line 516, in predict...
I evaluate the pretrained dinov2 with different decoder heads on KITTI Eigen Split in order to replicate the paper's numbers. I found the results much worse. Here's what I did....