Features and speed problem

[AbyszOne]

I like the script, but im missing 2 basic features: Black/white output with Contrast treshold Resolution output (Ex 512x512 center crop)

Then, maybe im doing something wrong. It detects CUDA, but its insanely slower vs other depth scripts using same models. Like 18s for 512x512 map in a 3060 using midas_small (1s in other scripts). Idk if is related, but it always say "process in 768" or 896, no matter what resolution image i use.

EDIT: I found BOOST as the problem. Now i get 3s for 512 map.

[graemeniedermayer]

The boost does hide the resolution output.

The "Clip and normalise" option might be close to the contrast threshold your thinking about. If you set the far clip and near clip to the same number it will output black and white images that cut at that value.

[AbyszOne]

Im trying to use renormalice for B/W output. It not replace forcing true B/W, but can work in some cases. Now, there is a trick for better depths (and general renders), wich is use enhance detail as preprocess. Here a simple example with same settings: ENHANCE STANDARD

Stronger lines and clarity leads to better maps overall. This isnt any difficult to implement as an option.

[AbyszOne]

The boost does hide the resolution output.

The "Clip and normalise" option might be close to the contrast threshold your thinking about. If you set the far clip and near clip to the same number it will output black and white images that cut at that value.

Yes and no in both cases. I get no resolution change using that sliders, just the original every time. Also, no crop selector. The clip option doenst replace a true b/w output, beside its indeed a "similar" function.

[graemeniedermayer]

Oh right those sliders just change the depth algorithms resolutions. If you run depthmap as a script in the img2img you can resize and crop the input image resolution (which will also change the output image resolutions). You might want to set denoising strength to zero. But I can see how you might want post-processing, which isn't currently included.

What you're saying about enhancing is interesting, but I think I'm missing context of what this particular enhance step means.

[AbyszOne]

I use my scripts to resize, but the point is that there is no such option, which should be elementary in image processing. About enhance, it's a trick I discovered creating my extension. Both denoise and enhance greatly affect either the SD generation or any analysis process. At the level where they should be fundamental options in GUIs. For example, a bilateral or median filter for noise, and another algorithm to add clarity and detail before rendering. Of course, this affects the depth map considerably too.