yolov8 可以将图片保存为 jpg

from ultralytics import YOLO
import cv2

# 加载模型
model = YOLO('yolov8n.pt')

# 预测（不自动保存）
results = model('input.jpg', save=False)

# 获取带标注的图片（numpy数组）
annotated_frame = results[0].plot()  # 绘制所有检测框

# 保存为JPG
cv2.imwrite('output.jpg', annotated_frame)

# 或转换为RGB后保存（OpenCV默认BGR）
from PIL import Image
rgb_frame = cv2.cvtColor(annotated_frame, cv2.COLOR_BGR2RGB)
Image.fromarray(rgb_frame).save('output_rgb.jpg', quality=95)

上面仅仅是实例代码，不能完整执行；

那么我们如何将这个结果通过uvc-gadget呢？

下载编译 uvc-gadget

编译 uvc-gadget程序

git clone https://gitlab.freedesktop.org/camera/uvc-gadget.git

我的代码是从这里下载的，编译方法参加这里的文档交叉编译

more cross-arm.txt 
[binaries]
c = 'aarch64-linux-gnu-gcc'
cpp = 'aarch64-linux-gnu-g++'
ar = 'aarch64-linux-gnu-ar'
windres = 'aarch64-linux-gnu-windres'
strip = 'aarch64-linux-gnu-strip'
exe_wrapper = 'wine64'

[host_machine]
system = 'linux'
cpu_family = 'x86_64'
cpu = 'x86_64'
endian = 'little'

编译后会生成

file build/src/uvc-gadget
build/src/uvc-gadget: ELF 64-bit LSB pie executable, ARM aarch64, version 1 (SYSV), dynamically linked, interpreter /lib/ld-linux-aarch64.so.1, BuildID[sha1]=373fb1f9f6f6d8cdaa8036dc05ad5751641fd1af, for GNU/Linux 3.7.0, with debug_info, not stripped

方法一

在板子上执行

./uvc-gadget uvc.0 -i output.jpg

但是这时候 output.jpg 不会自动更新需要修改程序

我们需要修改程序: uvc-gadget$ vim lib/jpg-source.c

static void jpg_source_fill_buffer(struct video_source *s,
                  struct video_buffer *buf)
{
   struct jpg_source *src = to_jpg_source(s);
   unsigned int size;

   reload_jpeg(s);
   /*
    * Nothing currently stops the user from providing a source file which is
    * larger than the buffer size calculated from the format we have set.
    * Clamp the size of the buffer we copy to be sure we don't overflow the
    * buffer we was allocated to receive it.
    */
   size = min(src->imgsize, buf->size);
   memcpy(buf->mem, src->imgdata, size);
   buf->bytesused = size;

   /*
    * Wait for the timer to elapse to ensure that our configured frame rate
    * is adhered to.
    */
   if (src->streaming)
       timer_wait(src->timer);
}

这里增加了 reload_jpeg函数，内容如下：

#define to_jpg_source(s) container_of(s, struct jpg_source, src)

static void reload_jpeg(struct video_source *s){
   struct jpg_source *src = to_jpg_source(s);

   int fd;
   int ret;

   if (src->imgdata) free(src->imgdata);

   fd = open(src->stream_path, O_RDONLY);
   if (fd == -1) {
   	printf("Unable to open MJPEG image '%s'\n", src->stream_path);
   	return ;
   }

   src->imgsize = lseek(fd, 0, SEEK_END);
   lseek(fd, 0, SEEK_SET);
   src->imgdata = malloc(src->imgsize);
   if (src->imgdata == NULL) {
   	printf("Unable to allocate memory for MJPEG image\n");
   	return ;
   }

   ret = read(fd, src->imgdata, src->imgsize);
   
   if (ret < 0) {
   	fprintf(stderr, "error reading data from %s: %d\n", src->stream_path, errno);
   	return ;
   }

   close(fd);

}

方法二、

需要修改 main.c 让其变成一个库不在是一个可执行文件把

vc-gadget$ ls lib/
compat      events.c      libcamera-source.cpp  meson.build-shared  slideshow-source.c  test-source.c  tools.h  uvc-formats.h  v4l2.c  v4l2-source.c    video-buffers.h
configfs.c  jpg-source.c  meson.build           mjpeg_encoder.cpp   stream.c            timer.c        uvc.c    uvc.h          v4l2.h  video-buffers.c  video-source.c

嵌入到项目

修改 src/main.c

让其入口main 函数变成一个线程入口，参数还是传递 uvc.0 -i output.jpg 这个时候这些都不变。

注意真的要放一个 output.jpg文件

修改 reload_jpg 函数

直接将 yolov8 输出的 jpg 数据不存放到硬盘，复制到 src 里即可。

src->imgdata, src->imgsize

如何将yolov8的图像通过 uvc-gadget输出

评论区 0