Record your screen, ship a demo. Free and open-source, GPU-accelerated, no watermarks, no subscriptions. Windows, macOS, Linux. Actively maintained.
TypeScript cross-platformelectronffmpegopen-sourceproduct-demorustscreen-capturescreen-recorderscreen-recordingscreen-studioscreencastspeech-to-texttypescriptvideo-editingvideo-editorwaylandwgpuwhisper
#pragma once
#include <vector>
// One capture format a camera advertises.
struct WebcamFormat {
int width = 0;
int height = 0;
int fps = 0;
/**
* True for MJPEG/H264 modes, which reach higher resolutions than a camera's
* uncompressed ones but cost a decode on the CPU for every frame.
*/
bool compressed = false;
};
// Picks the format to drive the camera at, given everything it advertises.
//
// Both capture backends used to leave this to the driver: Media Foundation was
// handed a media type with no MF_MT_FRAME_SIZE, and the DirectShow graph was
// rendered without touching IAMStreamConfig. In both cases the device answers
// with its *default* type, and for a UVC camera that is the first format it
// enumerates -- 640x480 even on a BRIO that offers 3840x2160. The recording was
// then upscaled into the overlay, which is the pixelation users reported.
//
// Rules, in order:
// * a compressed mode is out when decoding it costs more than 1080p at 30
// fps (an MJPEG 4K mode, say), unless the camera offers nothing cheaper.
// Under that ceiling it competes like any other mode: a C920 only reaches
// 1080p at 30 in MJPEG, its uncompressed 1080p runs at 5 fps;
// * only formats that fit inside the target in BOTH dimensions are eligible,
// so an ultrawide mode cannot sneak in on pixel count alone;
// * among those, the frame rate comes first: the highest rate up to the
// target wins, so a C920's 2304x1536 at 2 fps loses to its 1080p at 30.
// Ranking pixels first drove that camera at 2 to 5 fps on every preset;
// * then the most pixels;
// * then an uncompressed mode over a compressed one, which only adds a decode;
// * remaining ties go to the lowest frame rate that still reaches the
// target, else the highest available -- a 1080p60 mode is no help when we
// encode at 30;
// * if nothing fits (a camera that only does 4K), the smallest mode at the
// best rate wins, so we scale down rather than push 1 GB/s of RGB32
// through the encoder;
// * an empty list returns the target unchanged, letting the caller fall back
// to whatever the driver would have picked on its own.
WebcamFormat chooseWebcamFormat(
const std::vector<WebcamFormat>& available,
int targetWidth,
int targetHeight,
int targetFps);