The DVD upscaler is designed to re-encode videos to 720p.
go run cmd/client/* -h
go run cmd/client/* -path="[Path]"
go run cmd/client/* -dryRun -path="[Path]"
go run cmd/client/* -skipCleanup -skipDenoise -path="[Path]" -bitrate int
Maximum bitrate of the resulting video. (default 1950000)
-cores int
Number of CPU cores to use to encode the video. Defaults to one less than the total number of CPU cores.
-dryRun
Supply this flag when the video encoding step should be skipped.
-filter string
A regex value. Only scan movies whose title matches this value. (default ".*")
-force8Bit
Supply this flag when the resulting video's color depth should be 8-bit instead of 10-bit.
-forceAv1
Supply this flag when the resulting video's codec should be AV1 instead of HEVC.
-forceAvc
Supply this flag when the resulting video's codec should be AVC instead of HEVC.
-gop int
Maximum number of frames before forcing a keyframe. Larger values increase visual quality. (default 250)
-path string
The path to the directory to scan. (default "unknown")
-preset string
The preset to use. Slower preset values will produce better video quality. Valid preset values are: ultrafast superfast veryfast faster fast medium slow slower veryslow placebo (default "slow")
-skipCleanup
Supply this flag when the original videos should not be discarded.
-skipCrop
Supply this flag when letter-box bars in the source video should not be removed.
-skipDecomb
Supply this flag when interlaced video should not be converted to progressive video.
-skipDenoise
Supply this flag when the denoiser should not be used before scaling the video.
-skipNnedi
Supply this flag when the nnedi upscaler not be used to scale the video.
This application will scan the supplied directory path for subdirectories that contain a movie file whose name (less the media extension) exactly maches the subdirectory name. When found, the movie file will be copied to a local temporary directory, analyzed, and (if necessory) re-encoded.
In the case that the subdirectory does not contain a movie whose name exactly matches the subdirectory name this application will search the subdirectory for a movie file whose name (less the media extension) ends with - pt1. If a match is found this application will go into concatination mode which means that it will concatinate all videos in the subdirectory that end in - pt1 through - pt9000 into a single movie. Chapters names based on the concatinated movies will be added to the final movie to provide a convenient way to jump to the start of a specific concatinated video. The final movie will then be copied to a local temporary directory, analyzed, and (if necessory) re-encoded.
Because all media is copied to a local temporary directory the media optimizer is able to optimize remote directories that are mounted to your local filesystem. Thus, you can use Rclone to virturaly mount your remote cloud storage system to your local file system and then supply the path to this virtural mount to this application to optimize all the movies.
Server transcoding is when the media server that is hosting the file re-encodes the video in real-time while streaming the it to the client.
Transcoding a video file on the fly prevents more complex video codecs (such as HEVC), scaling algorithms (such as Nnedi), and denoisers (such as NLMeans) from being used effectively. Primitive scaling algorithms are undesirable because they produce a visual that has jaggier lines and more banding in gradients. Similarily, primitive denoisers are undesirable because they either cause the visual to look plasticy or retain more noise; the later requiring a higher bitrate to encode. Lastly, the older AVC video codec will need to be used or the HEVC video will need to be used with less advanced settings. In both cases, the video codec options for on the fly transcoding will require a higher bitrate to encode the video. Most of the time this bitrate is higher than the media server or video player's bitrate cap which forces the video codec to throw away needed bitrate which produces a blocky visual.
The server will trascode the video when the stored video file's resolution or maximum bitrate is higher than the server's or client's configured maximums. Additionally, the server will transcode the video when client's hardware does not support playback of the stored video's codec or bit-depth.
It is widely accepted that in most situations a video re-encoded using a CRF value of 16 is precieved as visually lossless compared to its source. Additionally, in most cases, the resulting average bitrate of a DVD re-encoded using the HEVC codec at a CRF of 16 with the slow preset will be under 2000kbps.
A couple of factors played into this decision. One reason is that many client video players default to a 2000kbps bandwidth cap. Staying underneath this cap will allow the media server to stream these movies to the client without transcoding the movie to a lower bitrate. Another reason is that most DVD movies have an average bitrate under 2000kbps when re-encoded using the HEVC codec at a CRF of 16 with the slow preset. This means that when proactively re-encoding a video the codec will only need to discard bitrate during the occasional high-movement scenes.
Many media servers and client video players default to a maximum resolution of 720p. Thus, we want to keep the maximum resolution to 720p or less to avoid server transcoding. Aditionally, proactively upscaling lower resolution videos to 720p allows us to use CPU intensive scaling algorithms such as Nnedi3 that can not be used in real-time.
When the source video has an 8-bit color depth re-encoding the video using a 10-bit color depth lets us achieve the same visual result for about 5% less bitrate. This lowers the chance that the re-encoding the video will hit the 2000kbps bitrate cap. Additionally, using the 10-bit color depth will prevent banding when the source material is a 10-bit gradient or the upscale algorithm and filters combine to produce a 10-bit gradient.
The HEVC codec is about 40% more efficient than the AVC codec which allows us to re-encode DVD videos using a CRF value of 16 to produce a movie that in most situations has a bitrate less than 2000kbps. Additioanlly, most consumer video decoders support the HEVC codec with a 10-bit color depth but only support the AVC codec with an 8-bit color depth. Thus, a media server is able to direct stream a 10-bit HEVC movie to the video player but must transcode the 8-bit AVC movie while streaming to the video player.
Strategic use of a denoiser allows us to dictate to the video codec what visuals to spend bitrate on by preemptively discarding bitrate used to encode video noise from the source video before re-encoding. NLMeans (a fairly powerful non-local means denoiser) is used ato denoise the video before scaling. After scaling Hqdn3d (a lightweight denoiser) and Unsharp (an image sharpener) are lightly used to remove temporal jitters that were introduced by the scaling process.
Cropping black bars from the video allows us to proactively scale the video to take advantage of the resolution that the black bars were occupying. This extra resolution helps the visual to avoid unnecessary jaggy lines and banding.