Administrator/opencv-MIRROR
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-09-11 04:43:22 -05:00
Code Issues Packages Projects Releases Wiki Activity

videoio: fix FFmpeg VideoCapture ignoring mid-stream pixel format change

Browse Source 
(cherry picked from commit f3c9241611cfb46dd17de2b3e3f25f6fe36b71f8)
This commit is contained in:
Lazizbek Ergashev
2026-08-12 15:07:57 +05:00
parent c6e84d494c
commit 28026fe2cf
2 changed files with 69 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
modules/videoio/src/cap_ffmpeg_impl.hpp
View File

@@ -582,6 +582,7 @@ struct CvCapture_FFMPEG
AVPacket packet;
Image_FFMPEG frame;
struct SwsContext *img_convert_ctx;
AVPixelFormat img_convert_ctx_format;
int64_t frame_number, first_frame_number;
@@ -648,6 +649,7 @@ void CvCapture_FFMPEG::init()
memset(&packet, 0, sizeof(packet));
av_init_packet(&packet);
img_convert_ctx = 0;
img_convert_ctx_format = AV_PIX_FMT_NONE;
avcodec = 0;
context = 0;
@@ -689,6 +691,7 @@ void CvCapture_FFMPEG::close()
{
sws_freeContext(img_convert_ctx);
img_convert_ctx = 0;
img_convert_ctx_format = AV_PIX_FMT_NONE;
}
if( picture )
@@ -1922,7 +1925,8 @@ bool CvCapture_FFMPEG::retrieveFrame(int flag, unsigned char** data, int* step,
if( img_convert_ctx == NULL ||
frame.width != video_st->CV_FFMPEG_CODEC_FIELD->width ||
frame.height != video_st->CV_FFMPEG_CODEC_FIELD->height ||
frame.data == NULL )
frame.data == NULL ||
(AVPixelFormat)sw_picture->format != img_convert_ctx_format )
{
#if LIBSWSCALE_BUILD >= CALC_FFMPEG_VERSION(6, 4, 100)
int buffer_width = video_st->CV_FFMPEG_CODEC_FIELD->width;
@@ -2000,6 +2004,8 @@ bool CvCapture_FFMPEG::retrieveFrame(int flag, unsigned char** data, int* step,
return false;
}
img_convert_ctx_format = (AVPixelFormat)sw_picture->format;
#if USE_AV_FRAME_GET_BUFFER
av_frame_unref(&rgb_picture);
rgb_picture.format = result_format;

62
modules/videoio/test/test_ffmpeg.cpp
View File

@@ -3,6 +3,7 @@
// of this distribution and at http://opencv.org/license.html.
#include "test_precomp.hpp"
#include "opencv2/imgcodecs.hpp"
using namespace std;
@@ -1188,4 +1189,65 @@ TEST(videoio_ffmpeg, seek_with_negative_dts)
}
}
// MJPEG stream whose chroma subsampling changes mid-stream at constant
// frame size must not reuse a conversion context built for the previous frame.
// related issue: https://github.com/opencv/opencv/issues/29699
TEST(videoio_ffmpeg, mjpeg_pixel_format_change)
{
if (!videoio_registry::hasBackend(CAP_FFMPEG))
throw SkipTestException("FFmpeg backend was not found");
Mat frame0(64, 64, CV_8UC3, Scalar(0, 0, 255));
Mat frame1(64, 64, CV_8UC3);
for (int y = 0; y < frame1.rows; y++)
for (int x = 0; x < frame1.cols; x++)
frame1.at<Vec3b>(y, x) = Vec3b((uchar)(x * 4), (uchar)(y * 4), (uchar)((x + y) * 2));
vector<uchar> buf0, buf1;
ASSERT_TRUE(imencode(".jpg", frame0, buf0,
{IMWRITE_JPEG_SAMPLING_FACTOR, IMWRITE_JPEG_SAMPLING_FACTOR_422, IMWRITE_JPEG_QUALITY, 100}));
ASSERT_TRUE(imencode(".jpg", frame1, buf1,
{IMWRITE_JPEG_SAMPLING_FACTOR, IMWRITE_JPEG_SAMPLING_FACTOR_420, IMWRITE_JPEG_QUALITY, 100}));
const string filename = tempfile("test_pixfmt_change.mjpeg");
std::ofstream file(filename.c_str(), ios::out | ios::trunc | std::ios::binary);
file.write(reinterpret_cast<char*>(buf0.data()), buf0.size());
file.write(reinterpret_cast<char*>(buf1.data()), buf1.size());
file.close();
VideoCapture cap(filename, CAP_FFMPEG);
ASSERT_TRUE(cap.isOpened());
Mat read0, read1;
ASSERT_TRUE(cap.read(read0));
ASSERT_TRUE(cap.read(read1));
Mat decodedRef0 = imdecode(buf0, IMREAD_COLOR);
Mat decodedRef1 = imdecode(buf1, IMREAD_COLOR);
ASSERT_FALSE(decodedRef0.empty());
ASSERT_FALSE(decodedRef1.empty());
ASSERT_EQ(read0.size(), decodedRef0.size());
ASSERT_EQ(read1.size(), decodedRef1.size());
// JPEG re-encoding is not bit exact, but a correctly decoded frame stays
// within a couple of intensity levels per channel; a stale conversion
// context (the bug) produces gross corruption of tens of levels.
const double maeThreshold = 5.0;
Mat diff0, diff1;
absdiff(read0, decodedRef0, diff0);
absdiff(read1, decodedRef1, diff1);
Scalar mae0 = mean(diff0);
Scalar mae1 = mean(diff1);
for (int c = 0; c < 3; c++)
{
EXPECT_LE(mae0[c], maeThreshold)
<< "First frame (channel " << c << ") decoded incorrectly";
EXPECT_LE(mae1[c], maeThreshold)
<< "Second frame (channel " << c << ") decoded incorrectly after pixel format change mid-stream";
}
remove(filename.c_str());
}
}} // namespace