Extracted Features

[Strike1999]

What a great work!

Could you provide the teachers' extracted features (e.g., SAM) for ImageNet? The extraction process is quite time-consuming for me due to limited computing resources.

Thanks.

[elicassion]

Hi @Strike1999

Unfortunately, SAM features are too large to upload (TB level for ImageNet-1k) :(

[Hermine-D]

Hi @Strike1999

Unfortunately, SAM features are too large to upload (TB level for ImageNet-1k) :(

@elicassion Can you provide the memory usage of SAM features? For example, 7TB or others. Because I'm not sure whether I have enough space to save features . Thank you very much.

[Strike1999]

Hi @Strike1999
Unfortunately, SAM features are too large to upload (TB level for ImageNet-1k) :(

Can you provide the memory usage of SAM features? For example, 7TB or others. Because I'm not sure whether I have enough space to save features . Thank you very much.

The features saved by Depth-Anything require 328GB, and the size of SAM features is 8 times that of Depth-Anything features, so I estimate the memory usage of SAM to be about 2.6TB.