💡 VisionDepth3D Feature Requests & Ideas!💡

[VisionDepth]

Hey everyone! 👋

First off, huge thanks for all the support and engagement with VisionDepth3D! The project has been getting a lot of attention, and I’d love to hear your thoughts on what features you'd like to see next.

💡 What ideas or improvements would you like to see?
Here are a few areas we could expand on:

• New 3D formats? (More output options for VR, better SBS/OU handling, etc.)
• Performance enhancements? (Faster depth map processing, batch optimizations)
• Custom depth models? (Support for more AI-based depth estimation models)
• Better UI/UX? (Improved interface, more controls, live preview, etc.)
• Post-processing options? (Sharpness tweaks, motion smoothing, Pulfrich effect adjustments)
• Something completely new? 🤔

If you have ideas, requests, or even just general feedback, drop them below! I want to make VisionDepth3D the best tool for high-quality 3D conversion, and your input will help shape its future. 🔥

Looking forward to hearing your thoughts!

[dylanwalker]

I would love to see support for 3DS formats on input. I know that may sound strange, but I would like to be able to leverage existing 3D footage to get a depth map.

[VisionDepth]

Hello,

Thanks for the suggestion, I'm confused on what format that is as I've never heard of it, give me some info on the format and I'll see if that's something that can be done

[lonnie776]

Is it possible to add the Video-Depth-Anything model?

[VisionDepth]

Hello,

Thanks for the suggestion, I tried to implement depthcrafter which is a stable diffusion model just like video depth anything and I couldn't get that working so the structure might be there to just swap over to video depth anything I'll work on it when I have time

[valimihoc-sys]

Hello,

Great piece of software, works great so far, really appreciate your hard work.
It would be cool if there was a image converter as well. Like batch depth generator for images, but also convert all of them them to sbs/vr.

[VisionDepth]

integrated a image generator in v3.7 in 3D tab select image in the input drop down, batch image will be added in next updates,

[3DVRFan]

Can we get a default naming scheme like "LRF_Full_SBS" at the end of finished files so VR headset programs etc can see it's a 3D file and display it as such.

[VisionDepth]

Added in v3.7

[C-Three-P-O]

Hi and thanx for the latest update. Made the interface more user friendly. I have been trying a lot of settings in IW3 but your program had (far) better results. Great job.

But then, some requests:

• Crop Black Bar: working would be nice. Without above/under the bar comes a lighter/whiter line. Cropping the bars with Handbrake creates a more cleaner image.

• Add more Base models: Small is nice for testing but from Base creates a watchable movie. I am using an TI Super 16GB. With Base (only one working is Any v2) i get around 10 FPS and with Large 4-5 FPS. I would like to try Dystill Any Dept Base, downloaded and named it according instructions but gives errors.

• IPD offset: what is the standard IPD offset in the -3D video creator- tab? And if scaling is enabled with how much it gets in/decreased?
  I have read the medial offset is 6.4 mm. Is this the standard value?

• Also the explanation of Depth Pop Gamma, Sharpness Strength and Convergence Strength contradicts (between in program and website)

[ValiMihawk]

Thanks for the 3d image generator. Any idea when the batch function can be implemented....one-by-one is kind of tedious.

[DeenDu02]

Hi, there are a few features I’d like to see added in future versions:

1. Support for importing and exporting high-quality formats such as ProRes, DNxHR, TIFF, EXE, DPX, and JPEG 2000.
2. Add the new repair model “M2SVid” as an option.
3. Could the 3D settings be modeled after the features in “Owl3D” and “Immersity”?

[sikejsudjek]

Please could you add half top bottom output.
Also is it possible to create separate output 24fps 1080p left and right files for frame packed 3d blu ray creation (is this already available)? In owl3d its called split SBS. I can then use the two files into sony vegas/dvd creator to create a 3d frame packed blu ray.

[VisionDepth]

Thanks for your interest in VD3D, T/B output is a legacy mode that I have taken out and plan to put back in once I get the sbs pipeline solid, when I took it out I was having trouble keeping same logic as one would need updating then the other would need usually causing t/b to break on me. There is Left/Right outputs as well as rendering both eyes and having separate files for each, in the encoder settings under load preview button, under output format change stereo output to the desired eye to be rendered

[sikejsudjek]

That's great thanks for that!

[Schniefel]

I think a working "estimated time" (depth map generation) would really help. On my system there were 2h estimated and it took roughly 2 days...

And i think performance, performance, performance. I really wondered why my RTX 4080 was on between 60-80% utilization. Tje VRAM was just on 4GB and I don't think that changing the batch size really did a difference. And i don't get why the value is not calculated automatically...

[VisionDepth]

Good point, and performance is something I want to keep improving.

In v4.1, depth generation is being moved into the shared Job Queue, so it should now show a clearer progress line with percentage, FPS, elapsed time, ETA, CPU, RAM, GPU, and VRAM usage.

The old ETA could be misleading because depth generation is not just one simple step. It is more like a chain. The app has to load the model, read the video frames, resize them, run the depth model, clean up the result, and write the output. If one part of that chain slows down, the ETA can change a lot.

For GPU and VRAM usage, 100% usage is not always expected. The GPU handles the depth model, but the CPU, RAM, disk, video decoding, and output writing are also involved. If one of those parts is slower, the GPU might sit at 60 to 80% even though it is still working properly.

VRAM is the same idea. If the selected model and resolution only need around 4 GB, then that is all it will use. Using more VRAM does not automatically mean better performance.

Batch size controls how many frames go into the model at once. A higher batch size can help in some cases, but it can also add more memory pressure. If the slowdown is from reading frames, resizing, writing output, or post-processing, changing batch size may not make much difference.

Auto batch size is a good idea, but I want to do it safely so it does not cause out-of-memory crashes on different GPUs.

Tooltips are also a great suggestion. I plan to bring those back into the program once the new v4.1 layout and theme system are fully settled. The User Guide link is being added under the Help menu too, but I agree that users should not have to constantly compare settings with GitHub. The app should explain more directly what each setting does.

[Schniefel]

I get the point with the VRAM usage. But when i don't see GPU and CPU maxed out, I would assume that the tasks for processing are not scheduled in an efficient way. I mean when i could run multiple instances of the software and when i process different part of the video i could probably achieve a more efficient way in processing.

[VisionDepth]

performance improvements are definitely still on my list.

[Schniefel]

Add tooltips to the various settings. It is a bit cumbersome to compare with the user guide on github.

[VisionDepth]

Thanks yes tooltips will be added back into the program, I wanted to get the whole pipeline to the new pyside6 UI and language files sorted out before I re add tooltips

[Schniefel]

Add a documentation for the pre-installed presets.
What is "Best", "Best_Settings", "Clean_Edge", ...?

[VisionDepth]

those are just starting points for users, they can select one see how it looks and play around with the settings if they want, also added a 3D assistant drop down that sets different depth personalities and users can use sliders to adjust those without having to get into the advanced sliders and they will have tool tips, (v5.0 release coming soon)

[Karlosspecial]

Unless I am missing something, would it be possible to get an auto aspect ratio or "same as original" in the 3D generator output and encoding? It seems that you have to select the aspect ratio manually

[VisionDepth]

when you go into processing options toggle preserve aspect ratio, the drop down in encoding settings is ignored i will put this toggle in the encoding settings because thats where it belongs, i just havent gotten there since i migrated to the new pyside6 UI

[mike413muo3]

Bonjour, serait-il possible de pouvoir sélectionner en sortie le format MVC pour ceux qui aime la conversion de film et préfère pouvoir les voir soit en 3d ou en 2D aujourd’hui je galère pour transformer mes full sbs en mvc merci pour votre prochaine réponse Olivier

[VisionDepth]

I will see if that is possible but I don't see why not, I will probably include this in v5.1 release as I am already in the Build phase of v5.0

[jdroads]

Installed and bought the CUDA version of v5 yesterday.

Is it possible to change the scroll wheel behavior when scrolling up and down in settings? if my mouse runs over a setting with a slider or number value, the value changes which means I have to be very mindful of where my mouse pointer is. I've screwed up settings a number of times and didn't know what the original settings were.

A better behavior would be to have the page/dialog only scroll up and down when the scroll wheel is used, and then if I want to change a value, click on the slider or value field to give it focus in order to change values.

I've struggled with this in the preview/data builds also and should have put in a request back then, but here I am.