search

CompVis / depth-fm

DepthFM: Fast Monocular Depth Estimation with Flow Matching
MIT License
395 stars 27 forks source link

Assumes CUDA, doesn't use safetensors #5

Open Vargol opened 8 months ago

Vargol commented 8 months ago

HI any chance you could change the checkpoint to be safetensor format, and maybe change the code not to assume the device is CUDA.

There are a lot of people that want to run stuff like this on MPS for example, and pickle files can contain executable code so should not be offered as a file format for security reasons.

FreakTheMighty commented 7 months ago

@Vargol is it possible to convert existing checkpoints or does that need to be retrained?

Vargol commented 7 months ago

Should be able to convert, safe tensors is just a way to store tensors without the possibility of embedded code running while loading. I'm not an expert though, but no one should be using formats that can execute code just by loading the files these days.