You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hey, I am trying to get the Dino features for an object using your interesting upsampling method.
When I run my custom code of Dinov2, I get the expected result (low resolution, without your upsampling).
But when using your model from the hub, I get different "colors" for the same object and I don't know what the problem is because you didn't release the full code.
Even if I mask the object to remove the background before your PCA, it doesn't help.
The semantics of the beer make sense but again the colors are different.
For example: the cap of the bottle has a different color than the rest of the bottle, but the color of the cap between different photos is different.
Example without masking: (light blue and strong blue for the same beer)
Any idea how to fix that?
Thanks for the great work!
The text was updated successfully, but these errors were encountered:
RoyAmoyal
changed the title
Same object and scene, different Upsampling and Dino features colors
Same object and scene, different Upsampling and Dinov2 features colors
Jul 1, 2024
Hey, I am trying to get the Dino features for an object using your interesting upsampling method.
When I run my custom code of Dinov2, I get the expected result (low resolution, without your upsampling).
But when using your model from the hub, I get different "colors" for the same object and I don't know what the problem is because you didn't release the full code.
Even if I mask the object to remove the background before your PCA, it doesn't help.
The semantics of the beer make sense but again the colors are different.
For example: the cap of the bottle has a different color than the rest of the bottle, but the color of the cap between different photos is different.
Example without masking: (light blue and strong blue for the same beer)
Any idea how to fix that?
Thanks for the great work!
The text was updated successfully, but these errors were encountered: