Struct Frame 

pub struct Frame {
    pub timestamp: Timestamp,
    pub size: Size,
    pub surface: Surface,
}

A decoded raw video frame: CPU I420, or a GPU frame when a hardware decoder produced one (NVDEC on Linux).

A GPU frame stays on the GPU until something needs bytes: feeding it to encode::Encoder::encode keeps it there (the zero-copy transcode path), while Surface::into_i420 downloads it.

Fields

timestamp: Timestamp

Presentation timestamp, carried through from the container. It rides out of the decoder with each picture, so a reordered frame (B-frames) keeps its own time rather than the input access unit’s.

size: Size

The decoded resolution, which is Config::resize when the backend honored it and the stream’s native size otherwise.

surface: Surface

Where the pixels live. Match on it for a zero-copy path, or call Surface::into_i420 for the universal one.

Implementations

impl Frame

pub fn resize(&self, size: Size) -> Result<Frame, Error>

A copy of this frame scaled to size (both dimensions even and non-zero), preserving the timestamp. A CUDA frame scales on the GPU (a box filter, correct at any downscale factor) and stays there, so resize -> encode never touches the CPU. Every other frame scales on the CPU, which for a CVPixelBuffer means a download first. When one output size is enough, prefer decoding straight to it (Config::resize), which is free on decoders with a hardware scaler; this method is for fanning one decoded stream out to several sizes.