~kris/hacks

sframe

02a1b32f6354d21c58f4d2bd92c608b2ec120336 — Kris Yotam 6 months ago 155c468
Add README
1 files changed, 116 insertions(+), 0 deletions(-)

A README.md
A README.md => README.md +116 -0
@@ 0,0 1,116 @@
# Simple Frame

A lightweight command-line tool for extracting unique frames from video
files. Built for research workflows where you need to dissect video
content into distinct, non-duplicate images with timestamps.

Simple Frame uses perceptual hashing to compare consecutive frames and
only saves those that are visually distinct, drastically reducing the
manual sorting required when working with extracted video stills.

Written in C following the [suckless philosophy](https://suckless.org/philosophy/).

## Usage

```
sframe [-v] [-t threshold] [-f png|jpg] [-o outdir] video [video...]
```

### Options

| Flag | Description | Default |
|------|-------------|---------|
| `-t N` | Hash threshold (0–64). Lower = stricter, fewer duplicates pass through. 0 removes only exact duplicates. | `8` |
| `-f fmt` | Output image format: `png` or `jpg` | `png` |
| `-o dir` | Output directory | `~/frames` |
| `-v` | Print version and exit | |

### Examples

Extract unique frames from a video:

```sh
sframe lecture.mp4
```

Use a stricter threshold with JPEG output:

```sh
sframe -t 4 -f jpg documentary.mkv
```

Process multiple videos into a custom directory:

```sh
sframe -o /tmp/research clip1.mp4 clip2.webm
```

## Output Structure

```
~/frames/
└── lecture/
    ├── lecture-00-00-01.234.png
    ├── lecture-00-00-15.678.png
    ├── lecture-00-01-02.345.png
    └── ...
```

Each video gets its own subdirectory. Frames are named as
`<video-name>-HH-MM-SS.mmm.<ext>` where the timestamp corresponds to
the frame's position in the video.

## How It Works

1. Decodes every frame of the video using FFmpeg
2. Scales each frame to 8x8 grayscale
3. Computes a 64-bit average perceptual hash
4. Compares the hamming distance against the previously saved frame's hash
5. Saves the frame only if the distance exceeds the threshold

This approach is fast and effective at eliminating duplicate and
near-duplicate frames (static shots, slow pans, talking heads with
minimal movement) while preserving genuine scene changes and visually
distinct content.

### Threshold Guide

| Value | Behavior |
|-------|----------|
| `0` | Only removes pixel-identical frames |
| `4` | Very strict — catches near-identical frames |
| `8` | Balanced — good default for most content |
| `16` | Loose — aggressive duplicate removal |
| `32+` | Very aggressive — only major scene changes |

## Dependencies

- FFmpeg libraries: `libavformat`, `libavcodec`, `libavutil`, `libswscale`
- `libpng`
- `libjpeg`

### Arch Linux

```sh
pacman -S ffmpeg libpng libjpeg-turbo
```

### Debian/Ubuntu

```sh
apt install libavformat-dev libavcodec-dev libavutil-dev libswscale-dev libpng-dev libjpeg-dev
```

## Building

```sh
make
sudo make install
```

Edit `config.mk` to change compile-time defaults (output format,
default threshold).

## License

MIT