Category: Encoding

InterFrame 2.6.0 released

This update adds a new option “FrameDouble” for users who always want to double the framerate.

For more details, please see the main post.

Changes since 2.5.2:

  • Added new boolean option “FrameDouble”, which when set to “true” will always double the framerate of the input video, unless NewNum is greater than 0.
  • Updated the “Recommended” script in the documentation

Why you should choose H-SBS over H-OU

You can either skip to the bottom of the article if you want to see the screenshot comparisons, or read through this to get the details behind the reasoning and testing methodology.

It’s common for 3D videos to be encoded in one of two formats: H-OU or H-SBS. These stand for Half Over-Under and Half Side-By-Side. In the former, the video for the left eye is stored above the video for the right eye, while in the latter the video for the left eye is stored to the left of the video for the right eye.

Getting H-SBS or H-OU instead of SBS or OU is the best option if you are going to watch the video over a network using DLNA, which is how most media servers work, since the full resolution won’t be used for full SBS or OU anyway. It will only display half of the pixels.

Additionally, a good encoder will have used a high quality resizer to halve the resolution in a H-OU or H-SBS file, meaning that the H-SBS/H-OU file will actually look higher quality when streamed over your network than a SBS/OU file – it’s unlikely that the media server and other hardware will use a slow, high-quality resizer while you play it, it will most likely use a quick, low quality one.

Why you should choose H-SBS over H-OU:
When you increase the resolution of an image – which is what software on your playback equipment (TV, monitor, etc.) needs to do to a H-SBS/H-OU video in order to stretch it back to its original size – the higher the starting resolution is, the better your resulting image will look. If you try to double the resolution of a 50×50 image to 100×100, the results will be inferior to doubling the same image from 1000×1000 to 2000×2000.
This may seem obvious. Of course a resizer will be able to be more accurate when it has more details to start with. Even though you are doubling the image in both situations, the situation that starts with the most details will result in the highest accuracy.

With H-OU, each eye sees a maximum resolution of 1920×540. This usually ends up being more like 1920×400, with the other 140 vertical pixels being the black bars at the top and bottom of the video frame. This gives us a great, full horizontal resolution, but a very small vertical resolution. In fact, this vertical resolution is even lower than DVDs, which can use up to 576 vertical pixels (typically between 404-484 for a cinematic movie).

By contrast, with H-SBS, each eye sees a maximum resolution of 960×1080, which usually ends up being more like 960×800 after the black bars are taken into account. This gives us a much more even selection of detail.

Both methods give us the same amount of pixels in total – 1036800 pixels per frame – but H-OU makes it harder for a resizer to enlarge the image as accurately, since it’s making the height so tiny. H-SBS halves the bigger number, resulting in two decent sets of rows and columns (960 and 800) instead of one large set of columns (1920) but one tiny set of rows (400).

Image examples:
It’s all well and good to talk about it, but seeing is more powerful.
For the following images, I’ve taken the original image (1920×800) and halved it using a bicubic (neutral) filter – the best choice for an image that will be enlarged later – then restored the original resolution using one of two resizing filters. One is Nearest Neighbor, which is a low quality filter, and the other is Bicubic Smooth, which is a high quality filter. Most TVs will have quality somewhere in between the two. My Panasonic VT60 seems to use something similar to Nearest Neighbor, which results in some very low quality moments with a H-OU source.

I’m using a website called Screenshot Comparison, which allows you to compare images by holding the mouse cursor over them, it’s a very convenient method.
For these comparisons, the H-OU image is visible when the cursor is not over the image, and the H-SBS image is visible when the cursor is over the image.

Image 1 using Nearest Neighbor
Image 1 using Bicubic Smooth
Image 2 using Nearest Neighbor
Image 2 using Bicubic Smooth

The most obvious parts to look at are the grills near the top of Image 1, the table near the bottom right of Image 1, and the details (lights, lines) on the ship in Image 2.

Important note: This article applies to monitors and TVs that use active 3D, not passive 3D. If your 3D glasses take batteries, plug in to recharge or have a button on them, then you use active 3D. Passive 3D TVs (like newer LG and Vizio TVs) should use H-OU, as widezu69 mentioned in the comments section below.

InterFrame 2.5.0 released

In addition to this release, I’ve also added new sample videos. These were made using the default InterFrame settings with GPU enabled (GPU=true).

For more details and the download link, please see the main post.

Changes since 2.4.0:

  • Minimized artifacts (thanks, Widezu69!)
  • Made motion smoother by default (only in GPU mode)
  • Updated libraries which improve scene-change detection, thanks to SVP

© 2023 Spirton

Theme by Anders NorénUp ↑