AmusementClub / vs-mlrt

Efficient CPU/GPU/Vulkan ML Runtimes for VapourSynth (with built-in support for waifu2x, DPIR, RealESRGANv2/v3, Real-CUGAN, RIFE, SCUNet and more!)
GNU General Public License v3.0
259 stars 15 forks source link

ESRGAN Anime_B6 #29

Open Jumaron opened 1 year ago

Jumaron commented 1 year ago

Hello guys,

i would love to ask if you could provide the ESRGAN Anime_B6 model as optimized .onnx Model like you have with the animvideov3 as example. I had problems converting it to .onnx, hope you could help.

As always thanks for the work and keep it up!

WolframRhodium commented 1 year ago

What's the problem?

Jumaron commented 1 year ago

Its not the fault of this project, it is most likely my code for transforming .pth to .onnx.

Jumaron commented 1 year ago

The error Code: [03/08/2023-14:44:46] [E] Error[4]: [network.cpp::nvinfer1::Network::validate::3076] Error Code 4: Internal Error (input: for dimension number 2 in profile 0 does not match network definition (got min=480, opt=480, max=480), expected min=opt=max=512).)

WolframRhodium commented 1 year ago

set dynamic_axes during onnx conversion

torch.onnx.export(..., dynamic_axes=dict(input={0: "batch", 2: "height", 3: "width"}))
Jumaron commented 1 year ago

Thanks, but now my 3070 GPU is running out of RAM i have 8 GB and i need 9 :( Still thanks, your a beast creating all of this.

dtlnor commented 1 year ago

Thanks, but now my 3070 GPU is running out of RAM i have 8 GB and i need 9 :( Still thanks, your a beast creating all of this.

you may try HolyWu's vs-realesrgan (no guarantee to be able to work with mlrt) https://github.com/HolyWu/vs-realesrgan/releases/tag/model

ViRb3 commented 1 year ago

You can get an onnx from here, I can confirm it works with mlrt: https://github.com/styler00dollar/VSGAN-tensorrt-docker/releases/tag/models

Jumaron commented 1 year ago

Thanks both of you! Working now as it should. <3