DeepEnhancer

[Selur]

BasicVSR++ is more suited to deal with compression and resizing artifacts, one of its strengthens is to repair chroma.
Spotless, DeDot and similar are more suited to remove such dirt

Don't like what DeepEnhancer does with the contrast.
For old clips, I usually prefer non-machine learning based solutions.

[zspeciman]

Selur, that was incredible work.  Thank you for the script, led me to discover why KillerSpots wasn't working for me previously, I renamed RemoveDirt.dll to RemoveDirtVS.dll to fix it

[Selur]

Oh, will upload a new dev to fix that.

[djilayeden]

Hi Selur,

Thanks for your feedback—totally understandable regarding DeepEnhancer’s contrast handling. It does tend to be a bit aggressive and might not be the best default choice for all clips, especially if a more subtle restoration is desired.

Still, I’ve found it surprisingly effective on very old footage, particularly from the 1920s–40s, where it manages to remove large blotches and stains that even Spotless and DeDot sometimes struggle with. I agree that BasicVSR++ is great for chroma repair and compression artifacts, but DeepEnhancer seems to go further in cleaning up physical damage, even if a bit heavy-handed.

In the end, it's great to have multiple tools available depending on the situation. The flexibility Hybrid and HAVC offer makes it possible to test and choose the best one for each clip—which is a huge strength!

Thanks again for your thoughts and for keeping everything so customizable.

Best regards

[Dan64]

If you like the sharpness, you can use FeMaSR

I attached a sample with the script.

To encode this script in Hybrid you must use as encoder NVenC and enable the option  DirectSynthLoading (check also MT) in NVEnc->Misc.

Hybrid by default uses vspipe, but in this case is not possible to use it.

I opened an issue on Vapoursynth project for this: vspipe don't filter out script messages

Dan

[Selur]

Didn't HolyWu fix that a while ago? (I encoded some stuff with BasicVSR++ and FeMaSR a few weeks ago without issue,.....)

[Dan64]

If you open the script attached in the previous post and run the preview are shown the following message.

Unable to import quantization op. Please install modelopt library (https://github.com/NVIDIA/TensorRT-Model...stallation) to add support for compiling quantized models
TensorRT-LLM is not installed. Please install TensorRT-LLM or set TRTLLM_PLUGINS_PATH to the directory containing libnvinfer_plugin_tensorrt_llm.so to use converters for torch.distributed ops
D:\Programs\Hybrid\64bit\Vapoursynth\Lib\site-packages\timm\models\layers\__init__.py:48: FutureWarning: Importing from timm.models.layers is deprecated, please import via timm.layers
warnings.warn(f"Importing from {__name__} is deprecated, please import via timm.layers", FutureWarning)

these message are not filteret-out by vspipe.

The problem is on vspipe. The pipe in output should provide only the video stream. 
It is a design bug and need to be fixed on vspipe side.

Dan

[Selur]

Ah must have done my BasicVSR++ FeMaSR encoding before I updated the torch-addon.
There is always something new.

[Selur]

It can be 'fixed' with:

import logging
import sys
logging.StreamHandler(sys.stderr)

=> Uploaded a new dev which adds this to the scripts if one of the add-ons is used in the script.

Cu Selur

[Dan64]

It can be 'fixed' with:

import logging
import sys
logging.StreamHandler(sys.stderr)

=> Uploaded a new dev which adds this to the scripts if one of the add-ons is used in the script.

Cu Selur

Added to the script Test-BasicVSR+FeMaSR.vpy the lines above and I put the script in input to Hybrid, still crashes.

Dan