Recently we finished a few narrative projects that were shot in 5K on the Red Epic but for various reasons were mastered in 2K. This was the plan from the outset and really showcases some of our favorite things about shooting 4K and mastering in 2K, namely the ability to reframe and stabilize shots in post as well as getting extra sharp looking results. However, at the end of the project there was some discussion about creating a 4K master. It’s not a trivial task considering that a significant amount of work would be needed to redo visual effects.
This is not the only use case for upresing; there are dozens of situations in which upresing HD and 2K footage to 4K is a tempting, potentially cost and time-saving solution. It’s not something we’ve done previously, so like most new things we approached from an analytical angle. With that, this HD to 4K Upres Shootout was conceived.
After some discussion we decided to lay out the following criteria for judging the upscaling solutions: First, visual quality is everything. Difficult to use interfaces or complicated workflows are things that we can learn and adapt to over time but we’ll have to live with the results of the upres solution we choose.
Secondly, there’s no such thing as a perfect upres. We’re not expecting the results to equal footage originally shot in 4K, but for comparison we’re including the actual 4K footage. At the very least it will show us how far away from perfect we are. It should be noted that ultimately a visually pleasing result is more desirable than one that is more accurate but with distracting artifacts.
Finally, processing time matters. While we noted that the interface or ease of use is not as important, significant render times would factor into our decision. This is also significant when considering if going back to the 4K original is a viable option from a time standpoint.
Throughout the test we took care to ensure that visual quality was maintained by using the same uncompressed 2K master for each application. The 4K footage comparison was from the original source (and hence some of the color grading appears different).
We decided to test a wide variety of options to really measure the current state of the industry. Here’s what we tested:
- After Effects Native Scaling
- After Effects offers a three levels of sampling used for it’s layers. These sampling methods are not proprietary and can be found in most image editing programs. We’re including them here so that they can be used to compare to unspecified methods in other applications. The nearest neighbor approach is the simplest. No interpolation of the pixel values is done and the new pixels are derived directly from the color values of which ever source pixel is closest in position. It’s not likely to give a very pleasing result, but it should be very fast. Second is bilinear sampling, which is the default for layers in After Effects. Bilinear sampling uses the values and relative positions of the four closest source pixels and interpolates a new pixel. Finally we’ll look at bicubic sampling, which like bilinear, samples neighboring pixels and interpolates a new value. Bicubic sampling uses a sample size of 16 pixels as opposed to bilinear’s 4 so it may provide smoother results at the cost of a little speed.
- Boris Continuum Complete UpRez
- UpRez is part of the Boris Continuum Complete suite of plugins available for various applications. We tested the After Effects version. UpRez uses a proprietary algorithm that has a few controls over quality and sharpness. For the test we adjusted the settings to look as good as possible over a wide range of frames.
- Red Giant Instant 4K
- Instant 4K is another upscaling solution that is available for Adobe After Effects and Premiere Pro. It uses a proprietary algorithm and also has some basic control over it’s quality and sharpness. For the test we adjusted the settings to look as good as possible over a wide range of frames.
- Cinnafilm Dark Energy Plug-In
- Dark Energy is a plug-in available for many transcoding suites and is based on the same engine that drives Dark Energy Professional. Dark Energy is used for video texture management including noise reduction and simulating film-like grain, so it’s no surprise that Dark Energy Plug-In states that is provides texture-aware scaling. Some options are available to adjust the strength of the noise-reduction and texturing. For the purposes of this test we left most of the settings in the “auto” mode.
- MTI UpRes
- MTI UpRes is part of MTI’s CORTEX Enterprise dailies suite. CORTEX provides a wide variety of image processing and media management features designed around the needs of a DIT. The Enterprise edition includes the UpRes feature, no doubt built around a proprietary algorithm.
- DVO Upscale
- DVO Upscale is available as a plug-in to DigitalVision’s suite of image processing and grading applications. We tested it using DigitalVision Phoenix, which is aimed at image restoration. Upscale is another plugin that makes use of a proprietary algorithm that makes a specific point to mention that it maintains high quality edges.