search

DepthAnything / Depth-Anything-V2

[NeurIPS 2024] Depth Anything V2. A More Capable Foundation Model for Monocular Depth Estimation
https://depth-anything-v2.github.io
Apache License 2.0
3.8k stars 323 forks source link

Size Mismatch for Depth-Anything-V2-Small Model #123

Open SuDaxia-kai opened 3 months ago

SuDaxia-kai commented 3 months ago

Hello,

I have recently downloaded the Depth-Anything-V2-Small model from the provided link on GitHub. According to the documentation, the model size is listed as 24.8M. However, upon downloading, I noticed that the actual size of the model does not match this specified size.

Here are the details:

Expected Size: 24.8M Actual Size: 99.2M Could you please verify if there is an error in the size listing or if there is another issue that I might be unaware of?

Thank you for your assistance.

LiheYoung commented 3 months ago

The 24.8M is the number of model parameters, while the 99.2M is the checkpoint storage size. Since our model is stored in float32 (4 bytes), the storage size correctly matches the parameter size.

SuDaxia-kai commented 3 months ago

The 24.8M is the number of model parameters, while the 99.2M is the checkpoint storage size. Since our model is stored in float32 (4 bytes), the storage size correctly matches the parameter size.

You can reply with:

"Thank you for the clarification, I understand now!"