search

Tencent / DepthCrafter

DepthCrafter: Generating Consistent Long Depth Sequences for Open-world Videos
https://depthcrafter.github.io
Other
636 stars 24 forks source link

Crashes memory on RTX 4090 Laptop (16Go) #10

Open nosy-b opened 2 weeks ago

nosy-b commented 2 weeks ago

Congrats for the great work ! I can't make it run on my laptop with RTX 4090 16Gb with the command _python run.py --video-path examples/example01.mp4 --max-res 512, it crashes with memory error.

Is there some other ways to optimize for small configuration? Thanks!

nosy-b commented 2 weeks ago

It runs with --max-res 256 but is there a way to support higher resolution ? (Even if it 's super slow) thanks :)

redman4585 commented 2 weeks ago

As it currently stands you will need a graphics card with more VRAM than a 4090 (They recommend a A100) to support the high resolution mode. With further optimizations that won't be necessary.

wbhu commented 2 weeks ago

It runs with --max-res 256 but is there a way to support higher resolution ? (Even if it 's super slow) thanks :)

Yes, you may feed cpu-offload to be "sequential", which can save a lot of memory, but slow. Refer to https://github.com/Tencent/DepthCrafter/blob/main/run.py#L37