How to render images to /dev/video0 using v4l2loopback? - c++

Question

I've been trying to render images to /dev/video. I can get something to sort of display but it's somewhat scrambled.
I first started off trying to render a normal RGB24 image (based off this example https://stackoverflow.com/a/44648382/3818491), but the result (below) was a scrambled image.
#include <stdio.h>
#include <fcntl.h>
#include <errno.h>
#include <unistd.h>
#include <iostream>
#include <sys/ioctl.h>
#include <linux/videodev2.h>
#include <CImg.h>
#define VIDEO_OUT "/dev/video0" // V4L2 Loopack
#define WIDTH 1280
#define HEIGHT 720
int main() {
using namespace cimg_library;
CImg<uint8_t> canvas(WIDTH, HEIGHT, 1, 3);
const uint8_t red[] = {255, 0, 0};
const uint8_t purple[] = {255, 0, 255};
int fd;
if ((fd = open(VIDEO_OUT, O_RDWR)) == -1) {
std::cerr << "Unable to open video output!\n";
return 1;
}
struct v4l2_format vid_format;
vid_format.type = V4L2_BUF_TYPE_VIDEO_OUTPUT;
if (ioctl(fd, VIDIOC_G_FMT, &vid_format) == -1) {
std::cerr << "Unable to get video format data. Errro: " << errno << '\n';
return 1;
}
size_t framesize = canvas.size();
int width = canvas.width(), height = canvas.height();
vid_format.fmt.pix.width = width;
vid_format.fmt.pix.height = height;
vid_format.fmt.pix.pixelformat = V4L2_PIX_FMT_RGB24;
vid_format.fmt.pix.sizeimage = framesize;
vid_format.fmt.pix.field = V4L2_FIELD_NONE;
if (ioctl(fd, VIDIOC_S_FMT, &vid_format) == -1) {
std::cerr << "Unable to set video format! Errno: " << errno << '\n';
return 1;
}
std::cout << "Stream running!\n";
while (true) {
canvas.draw_plasma();
canvas.draw_rectangle(
100, 100, 100 + 100, 100 + 100, red, 1);
canvas.draw_text(5,5, "Hello World!", purple);
canvas.draw_text(5, 20, "Image freshly rendered with the CImg Library!", red);
write(fd, canvas.data(), framesize);
}
}
So I checked what (I think) /dev/video expects which seems to be YUV420P.
v4l2-ctl --list-formats-ext 130 ↵
ioctl: VIDIOC_ENUM_FMT
Type: Video Capture
[0]: 'YU12' (Planar YUV 4:2:0)
Size: Discrete 1280x720
Interval: Discrete 0.033s (30.000 fps)
So I attempted to convert the frame that format (using this code to quickly test).
Adjusting the spec to:
vid_format.fmt.pix.width = width;
vid_format.fmt.pix.height = height;
vid_format.fmt.pix.pixelformat = V4L2_PIX_FMT_YUV420;
vid_format.fmt.pix.sizeimage = width*height*3/2; // size of yuv buffer
vid_format.fmt.pix.field = V4L2_FIELD_NONE;
That results in this (which seems to be from what I've gathered the structure of a yuv420 image but still rendered incorrectly).
What does /dev/video0 expect?

Answer 1

After a lot of hacking around, I've managed to generate a valid YUYV video/image to send to /dev/video0.
First I make a buffer to hold the frame:
// Allocate buffer for the YUUV frame
std::vector<uint8_t> buffer;
buffer.resize(vid_format.fmt.pix.sizeimage);
Then I write the current canvas to the buffer in YUYV format.
bool skip = true;
cimg_forXY(canvas, cx, cy) {
size_t row = cy * width * 2;
uint8_t r, g, b, y;
r = canvas(cx, cy, 0);
g = canvas(cx, cy, 1);
b = canvas(cx, cy, 2);
y = std::clamp<uint8_t>(r * .299000 + g * .587000 + b * .114000, 0, 255);
buffer[row + cx * 2] = y;
if (!skip) {
uint8_t u, v;
u = std::clamp<uint8_t>(r * -.168736 + g * -.331264 + b * .500000 + 128, 0, 255);
v = std::clamp<uint8_t>(r * .500000 + g * -.418688 + b * -.081312 + 128, 0, 255);
buffer[row + (cx - 1) * 2 + 1] = u;
buffer[row + (cx - 1) * 2 + 3] = v;
}
skip = !skip;
}
Note:
CImg has RGBtoYUV has an in-place RGB to YUV conversion, but for some reason calling this on a uint8_t canvas just zeros it.
It also has get_YUVtoRGB which (allocates and) returns a CImg<float> canvas, which I think you multiply each value by 255 to scale to a byte, however, whatever I tried that did not give the correct colour. Edit: I likely forgot the +128 bias (though I still prefer not reallocating for each frame)
My full code is here (if anyone wants to do something similar) https://gist.github.com/MacDue/36199c3f3ca04bd9fd40a1bc2067ef72