purnasai / dino_v2 Goto Github PK

Hello! Thank you for sharing the great code. I have a question. Could you please explain why you applied 'transformer.norm' in the 'forward' function of 'DinoVisionTransformerClassifier'?

def forward(self, x):
    x = self.transformer(x)
    x = self.transformer.norm(x)
    x = self.classifier(x)
    return x

Hi, your repo is really helpful. I have one minor question, how do you know this threshold pca_features_bg = pca_features[:, 0] > 0.35 in third-party/Dino_V2/2.PCA_visualization.ipynb IN [10]? I am not sure how to infer this number from the first histogram. Could you please explain it?

Moreover, according to the original paper of DINOv2, it says Background is removed by removing patches with a negative score of the first PCA component. (Figure 9 caption) May I ask what is the relationship between the negative score and your chosen threshold. Many thanks!

Hello,

Thank you for sharing your great article on Medium and GitHub. I was inspired a lot.
In 3.DinoV2_VS_ResnetClassification.ipynb, you load the dinov2_vits14 model, and I don't see anywhere you freeze the parameters in the dinov2.

Does it mean that during the training process, you tune all the parameters in the DinoVisionTransformerClassifier (including transformer and linear layer)?
Then does it mean that all the benefits come from the pretrained model and curated data is discarded?

Can you add requirements.txt file