[ENHANCEMENT] Ignore or Post-Cut Full Transparency When Creating Depthmaps

[shinra358]

Real 3D in 2D game. Only the dragon is in 3D right now. And the dual snouts is because the game draws 2 enemies on the screen for 1 enemy and i didnt target the doubled enemy yet. butcha get the gist. Depth map has to be created for every asset and it gets injected like texture injection. Now ff6 can be played in 3D like FF1 for nintendo 3DS. Use crosseye method or any 3D headset to view.

^What I'm asking is for iw3 to auto cut out the full transparent areas when you are getting the depthmap of images like the above picture. This reduces workload for me when creating these depthmaps to generate 3D in 2D based games.

[nagadomi]

I have tried this before and found the following:

1. PIL does not support exporting 16 bit grayscale images with an alpha channel, so it is not possible to add alpha to depth files.
2. Filling transparent areas in the depth map using the alpha channel causes severe artifacts during iw3 warping. This is effectively equivalent to disabling Depth Antialias and Edge Dilation. For this reason, it is better for iw3 not to apply the alpha channel to the depth map.
3. Filling transparent areas in the RGB image with a solid color can sometimes cause distortions in the generated depth map.

Regarding point 2, this may not be an issue since you intend to use the depth map in a different application.
Ultimately, I think the best approach is for your software to use the RGB image's alpha channel to mask the depth map.

[nagadomi]

As mentioned above, depth images with an alpha channel cannot be exported. However, since there is also interest in this feature as discussed in #279 , I will add support for RGBA export and import.

[nagadomi]

Added in #704

[shinra358]

i pressed update.bat, update_installer.bat, and then update.bat again. I don't see the option and the export didnt have transparency. Is it live or still waiting?

[nagadomi]

If you are using the dev branch, this should already be applied. As mentioned above, depth maps do not have an alpha channel, but fully transparent regions are filled with black.

[shinra358]

Oh i don't have dev branch. I'll wait til it comes to main. But after that can you:

Alpha Thresholding
3D viewers and displacement maps do not handle semi-transparent depth values well. This approach removes the fully transparent background but forces any semi-transparent pixels to become fully opaque on the depth map.
Result, Fully transparent pixels become fully transparent on the depth map. Semi-transparent pixels become 100% solid/opaque.

Filters used: alphaextract to pull the mask, a lut (Look-Up Table) filter to clamp any alpha value greater than 0 up to 255, and alphamerge to apply the modified mask to the depth map.

With ffmpeg: ffmpeg -i original.png -i depthmap.png -filter_complex "[0:v]alphaextract,lut=y='sgn(val)*255'[mask];[1:v][mask]alphamerge" -c:v png output.png

Then add as an option to make DepthMap with Transparent BG to allow the program to do this post conversion so we can do this in one swoop. The easy button way.

[nagadomi]

For depth maps, semi transparent regions are already treated as fully opaque. The alpha channel is effectively a binary mask where only alpha = 0 is transparent and everything else is opaque.
If the alpha boundary appears blurred, it is caused by Edge Dilation or Depth AA. Set them to 0 to disable them.

For RGB images, the original alpha channel is preserved.

[shinra358]

the focal point is real transparency in the depth map. not black bg, just transparency. i dunno if you understood that because the response is speaking of semi transparency only. and the previous response is speaking of black representing the transparency. but the ffmpeg code i gave is using the original image as a mask to delete the black representing transparency after the conversion so the black background turns real transparent in the final image.