search

AiuniAI / Unique3D

[NeurIPS 2024] Unique3D: High-Quality and Efficient 3D Mesh Generation from a Single Image
https://wukailu.github.io/Unique3D/
MIT License
3.11k stars 246 forks source link

onnx error #55

Open shudct opened 5 months ago

shudct commented 5 months ago

[ONNXRuntimeError] : 6 : RUNTIME_EXCEPTION : Non-zero status code returned while running Conv node. Name:'/conv_up2/Conv' Status Message: /onnxruntime_src/onnxruntime/core/framework/bfc_arena.cc:376 void onnxruntime::BFCArena::AllocateRawInternal(size_t, bool, onnxruntime::Stream, bool, onnxruntime::WaitNotificationFn) Available memory of 898621440 is smaller than requested bytes of 2147631360\n

I make a API demo, for the first several post, it success. however, onnx report a error later. It seems OOM. Is there any solution?

wukailu commented 4 months ago

Looks like it's low on video memory, what's your GPU video memory? Recommended 18GB or more

liubc-ai commented 4 months ago

I test on RTX3090 with 24G VRAM, and I meet the same problem when use realesrgan to upsample. Controling the input image size < 512 can avoid this problem

shudct commented 4 months ago

Looks like it's low on video memory, what's your GPU video memory? Recommended 18GB or more

I use the 4090 with 24G memory. I notice that error only when I pass image < 512. image image The second "run_sr_fast" make the code OOM. I guess after the first "run_sr_fast" process, the onnx memory is not release, but I don't know how to solve it.

wukailu commented 4 months ago

Oh, I noticed yesterday that this might be an issue. If the image resolution is below 512, it will be upsampled by 4 times to reach a resolution of 2048. This may cause a memory overflow on some very wide images.