Georg Pelz

Small update: * vsfemasr 2.0.0 works fine. * got BasicVSR++ working with [site-packages_mmcv_p312_cuda121.zip](https://forum.selur.net/attachment.php?aid=2531) provided by Dan64. (ugly since it's not a wheel, but it works ;)) so atm. only: *...

So the shader needs to me rewritten to assume YUV input to work with placebo.Shader or is there a mistake in the shader that it doesn't work with RGB as...

same error here

Instead of having to add a 'cuda-safe', 'qsv-save',... option adding an additional parameter i.e. 'fallback: bool, whether to fallback to cpu encoding on error' seems like a better way to...

Sure, the main difference is whether this is implemented once or whether each user has to make sure he uses the wrapper. :) So would be nice if this could...

Sounds good to me, if the decoder should work and crashes it should crash. If no decode i.e. ProRes, Huffyuv,... exists falling back to cpu sound good.

Thanks for the update.

Can this be done used with [vsmlrt.inference(..)](https://github.com/AmusementClub/vs-mlrt/blob/ae2ceaffe052398c508941b56fb8319a4265c646/scripts/vsmlrt.py#L3080) the same way? i.e. `clip = vsmlrt.inference([clip[:-2], clip[1:-1], clip[2:],network_path="", backend=Backend.TRT(fp16=True,device_id=0,bf16=False,num_streams=1,verbose=True,use_cuda_graph=True,workspace=1073741824,builder_optimization_level=3,engine_folder="J:/TRT"))` (I like the convenience of the inference function. :)) for N temporal frames,...

Nice! 🥇 but,... using TRT: `clip = vsmlrt.inference([clip[i:-(3 - i - 1) or None] for i in range(3)],network_path="C:/Users/Selur/Desktop/testing/.onnx",overlap=(64, 64),tilesize=[384, 576], backend=Backend.TRT(fp16=True,device_id=0,bf16=False,num_streams=1,verbose=True,use_cuda_graph=False,workspace=1073741824,builder_optimization_level=3,engine_folder="J:/TRT")) # 720x576` It fails for me with: ``` clip...

> 4.15 lite v2, 4.17 lite v2, 4.17 worth checking for non-anime content? (so far 4.18 seemed to prefer best for me)