mirror of
https://github.com/levogevo/ffmpeg-av1-builder.git
synced 2026-01-15 16:56:18 +00:00
111 lines
6.9 KiB
Markdown
111 lines
6.9 KiB
Markdown
# This project has been archived. Check here for the updated version: https://github.com/levogevo/ffmpeg-builder
|
|
|
|
|
|
|
|
|
|
# ffmpeg-av1-builder
|
|
|
|
## General
|
|
This repository is a simple collection of bash scripts. Scripts have been tested on:
|
|
- Ubuntu/Debian/Arch x86_64
|
|
- Ubuntu aarch64
|
|
- Windows using WSL Ubuntu/Debian
|
|
- MacOS 15 on Mac Mini M4
|
|
|
|
If you are not on any of these platforms, you will need to setup dependencies yourself.
|
|
With the scripts, you can:
|
|
|
|
1. Install required dependencies using `./scripts/install_deps.sh`
|
|
2. Build and install ffmpeg from source using `./scripts/build.sh`
|
|
3. Install an encoding script using `./scripts/recc_encode.sh -I`
|
|
4. Install a film-grain estimation script using `./scripts/estimate_fg.sh -I`
|
|
5. Benchmark the different encoders using `./scripts/benchmark.sh`
|
|
|
|
## Encode script
|
|
The installation of the encoding script creates a symlink to this repo's `./scripts/recc_encode.sh` so do NOT remove this repo or the functionality of `encode` will FAIL. The `encode` script is a very simple way to use SvtAv1-PSY for video and Opus for audio for encoding, which are arguably the most ideal encoders as of writing this. `encode` does the following:
|
|
- Maps all streams except image streams (effectively no poster/image in the output)
|
|
- Sets the audio bitrate at 64kbps per channel for each audio track
|
|
- Formats the audio output channel layout correctly for Opus
|
|
- Attempts to encode with Dolby Vision and encodes without if the attempt fails
|
|
- Preset 3 and CRF 25 for for SvtAv1
|
|
- Enables sane advanced parameters
|
|
- Removes video stream title name from output
|
|
- Adds track statistics for proper output video/audio bitrate reporting
|
|
Read the specifics in the actual file : `./scripts/recc_encode.sh`
|
|
|
|
```bash
|
|
encode -i input_file [options]
|
|
[-p] print the command instead of executing it
|
|
[-c] use cropdetect
|
|
[-s] use same container as input, default is mkv
|
|
[-v] Print relevant version info
|
|
[-g NUM] set film grain for encode
|
|
|
|
[output_file] if not set, will create at $HOME/
|
|
|
|
[-I] Install this as /usr/local/bin/encode
|
|
[-U] Uninstall this from /usr/local/bin/encode
|
|
```
|
|
Example usage:
|
|
- `encode -i input.mkv output.mkv` standard usage
|
|
- `encode -i input.mkv -p true` prints out what it will do, does not start encoding
|
|
- `encode -i input.mkv` no output filename will create an output video in your home folder (~/)
|
|
- `encode -i input.mkv -g 20` will encode with film-grain synthesis=20 WITH denoising
|
|
|
|
## Estimate film grain script
|
|
The installation of the script creates a symlink to this repo's `./scripts/estimate_fg.sh` so do NOT remove this repo or the functionality of `estimate-film-grain` will FAIL. The `estimate-film-grain` script is a way to estimate the ideal film grain of a video by encoding it at different film grain values and observing at what point does a higher film grain value result in diminishing returns.
|
|
```bash
|
|
estimate-film-grain -i input_file [options]
|
|
[-o output_file] file to output results to
|
|
[-l NUM] low value to use as minimum film-grain
|
|
[-s NUM] step value to use increment from low to high film-grain
|
|
[-h NUM] high value to use as maximum film-grain
|
|
[-p] plot bitrates using gnuplot
|
|
|
|
[-I] Install this as /usr/local/bin/estimate-film-grain
|
|
[-U] Uninstall this from /usr/local/bin/estimate-film-grain
|
|
```
|
|
Example usage:
|
|
- `estimate-film-grain -i input.mkv -p` plot input.mkv
|
|
```
|
|
1 +------------------------------------------------------------------------------------------------------+
|
|
| *****G***** + + + |
|
|
| *****G** '/tmp/plot.dat' ***G*** |
|
|
0.95 |-+ ***** +-|
|
|
| **G* |
|
|
| *** |
|
|
| **** |
|
|
0.9 |-+ *G* +-|
|
|
| **** |
|
|
| **** |
|
|
0.85 |-+ *G* +-|
|
|
| *** |
|
|
| **** |
|
|
0.8 |-+ *G* +-|
|
|
| *** |
|
|
| **** |
|
|
| *G* |
|
|
0.75 |-+ *** +-|
|
|
| **** |
|
|
| *G* |
|
|
0.7 |-+ *** +-|
|
|
| ** |
|
|
| *** |
|
|
| *G* |
|
|
0.65 |-+ *** +-|
|
|
| **** |
|
|
| + + + *|
|
|
0.6 +------------------------------------------------------------------------------------------------------+
|
|
```
|
|
## Benchmark script
|
|
AV1 encode quality is tested against 5 different open source videos using libsvtav1, librav1e, and libaom.
|
|
Netflix's libvmaf is used to analyze quality of the encodes against the original files.
|
|
|
|
Output after running `./scripts/benchmark.sh` is in `./benchmark/results.txt` which contains the following values for each encode:
|
|
|
|
* the time taken
|
|
* psnr_hvs
|
|
* cambi
|
|
* float_ms_ssim
|
|
* vmaf
|