Using ffmpeg-encoded video from javacv on Android causes native code to crash… here is a solution to the problem.
Using ffmpeg-encoded video from javacv on Android causes native code to crash
Note: I’ve updated this since I originally posed this issue to reflect some of what I’ve learned in loading live camera images into the ffmpeg library.
I’m using ffmpeg in javacv compiled for Android to encode/decode video for my app. (Note that initially, I tried ffmpeg-java, but it has some incompatible libraries).
Original problem: The problem I’m having is that I’m currently using every frame as a Bitmap (just a plain android.graphics. bitmap) and I don’t know how to populate it into the encoder.
Solution in ffmpeg for javacv: With avpicture_fill(), the Android format should be YUV420P, although I can’t verify until my encoder issue (below) is resolved.
avcodec.avpicture_fill((AVPicture)mFrame, picPointer, avutil. PIX_FMT_YUV420P, VIDEO_WIDTH, VIDEO_HEIGHT)
Now the problem: the thread that should have actually encoded the data crashed. I get a big native code stack trace that I can’t understand. Does anyone have any suggestions?
This is the code I used to instantiate all ffmpeg libraries :
avcodec.avcodec_register_all();
avcodec.avcodec_init();
avformat.av_register_all();
mCodec = avcodec.avcodec_find_encoder(avcodec. CODEC_ID_H263);
if (mCodec == null)
{
Logging.Log("Unable to find encoder.");
return;
}
Logging.Log("Found encoder.");
mCodecCtx = avcodec.avcodec_alloc_context();
mCodecCtx.bit_rate(300000);
mCodecCtx.codec(mCodec);
mCodecCtx.width(VIDEO_WIDTH);
mCodecCtx.height(VIDEO_HEIGHT);
mCodecCtx.pix_fmt(avutil. PIX_FMT_YUV420P);
mCodecCtx.codec_id(avcodec. CODEC_ID_H263);
mCodecCtx.codec_type(avutil. AVMEDIA_TYPE_VIDEO);
AVRational ratio = new AVRational();
ratio.num(1);
ratio.den(30);
mCodecCtx.time_base(ratio);
mCodecCtx.coder_type(1);
mCodecCtx.flags(mCodecCtx.flags() | avcodec. CODEC_FLAG_LOOP_FILTER);
mCodecCtx.me_cmp(avcodec. FF_LOSS_CHROMA);
mCodecCtx.me_method(avcodec. ME_HEX);
mCodecCtx.me_subpel_quality(6);
mCodecCtx.me_range(16);
mCodecCtx.gop_size(30);
mCodecCtx.keyint_min(10);
mCodecCtx.scenechange_threshold(40);
mCodecCtx.i_quant_factor((float) 0.71);
mCodecCtx.b_frame_strategy(1);
mCodecCtx.qcompress((float) 0.6);
mCodecCtx.qmin(10);
mCodecCtx.qmax(51);
mCodecCtx.max_qdiff(4);
mCodecCtx.max_b_frames(1);
mCodecCtx.refs(2);
mCodecCtx.directpred(3);
mCodecCtx.trellis(1);
mCodecCtx.flags2(mCodecCtx.flags2() | avcodec. CODEC_FLAG2_BPYRAMID | avcodec. CODEC_FLAG2_WPRED | avcodec. CODEC_FLAG2_8X8DCT | avcodec. CODEC_FLAG2_FASTPSKIP);
if (avcodec.avcodec_open(mCodecCtx, mCodec) == 0)
{
Logging.Log("Unable to open encoder.");
return;
}
Logging.Log("Encoder opened.");
mFrameSize = avcodec.avpicture_get_size(avutil. PIX_FMT_YUV420P, VIDEO_WIDTH, VIDEO_HEIGHT);
Logging.Log("Frame size - '" + mFrameSize + "'.");
mPic = new AVPicture(mPicSize);
mFrame = avcodec.avcodec_alloc_frame();
if (mFrame == null)
{
Logging.Log("Unable to alloc frame.");
}
This is what I hope to be able to do next :
BytePointer picPointer = new BytePointer(data);
int bBuffSize = mFrameSize;
BytePointer bBuffer = new BytePointer(bBuffSize);
int picSize = 0;
if ((picSize = avcodec.avpicture_fill((AVPicture)mFrame, picPointer, avutil. PIX_FMT_YUV420P, VIDEO_WIDTH, VIDEO_HEIGHT)) <= 0)
{
Logging.Log("Couldn't convert preview to AVPicture (" + picSize + ")");
return;
}
Logging.Log("Converted preview to AVPicture (" + picSize + ")");
VCAP_Package vPackage = new VCAP_Package();
if (mCodecCtx.isNull())
{
Logging.Log("Codec Context is null!");
}
encode the image
int size = avcodec.avcodec_encode_video(mCodecCtx, bBuffer, bBuffSize, mFrame);
int totalSize = 0;
while (size >= 0)
{
totalSize += size;
Logging.Log("Encoded '" + size + "' bytes.");
Get any delayed frames
size = avcodec.avcodec_encode_video(mCodecCtx, bBuffer, bBuffSize, null);
}
Logging.Log("Finished encoding. (" + totalSize + ")");
However, as of now, I don’t know how to put the bitmap in the right section, or if I set that setting correctly.
A few notes about the code:
– VIDEO_WIDTH = 352
– VIDEO_HEIGHT = 288
– VIDEO_FPS = 30;
Solution
After a lot of searching, I found that you have to load pointers in a rather strict and clumsy way. That’s how I made everything work :
Codec settings:
avcodec.avcodec_register_all();
avcodec.avcodec_init();
avformat.av_register_all();
/* find the H263 video encoder */
mCodec = avcodec.avcodec_find_encoder(avcodec. CODEC_ID_H263);
if (mCodec == null) {
Log.d("TEST_VIDEO", "avcodec_find_encoder() run fail.");
}
mCodecCtx = avcodec.avcodec_alloc_context();
picture = avcodec.avcodec_alloc_frame();
/* put sample parameters */
mCodecCtx.bit_rate(400000);
/* resolution must be a multiple of two */
mCodecCtx.width(VIDEO_WIDTH);
mCodecCtx.height(VIDEO_HEIGHT);
/* frames per second */
AVRational avFPS = new AVRational();
avFPS.num(1);
avFPS.den(VIDEO_FPS);
mCodecCtx.time_base(avFPS);
mCodecCtx.pix_fmt(avutil. PIX_FMT_YUV420P);
mCodecCtx.codec_id(avcodec. CODEC_ID_H263);
mCodecCtx.codec_type(avutil. AVMEDIA_TYPE_VIDEO);
/* open it */
if (avcodec.avcodec_open(mCodecCtx, mCodec) < 0) {
Log.d("TEST_VIDEO", "avcodec_open() run fail.");
}
/* alloc image and output buffer */
output_buffer_size = 100000;
output_buffer = avutil.av_malloc(output_buffer_size);
size = mCodecCtx.width() * mCodecCtx.height();
picture_buffer = avutil.av_malloc((size * 3) / 2); /* size for YUV 420 */
picture.data(0, new BytePointer(picture_buffer));
picture.data(1, picture.data(0).position(size));
picture.data(2, picture.data(1).position(size / 4));
picture.linesize(0, mCodecCtx.width());
picture.linesize(1, mCodecCtx.width() / 2);
picture.linesize(2, mCodecCtx.width() / 2);
Process preview data:
//(1)Convert byte[] first
byte[] data420 = new byte[data.length];
convert_yuv422_to_yuv420(data, data420, VIDEO_WIDTH, VIDEO_HEIGHT);
(2) Fill picture buffer
int data1_offset = VIDEO_HEIGHT * VIDEO_WIDTH;
int data2_offset = data1_offset * 5 / 4;
int pic_linesize_0 = picture.linesize(0);
int pic_linesize_1 = picture.linesize(1);
int pic_linesize_2 = picture.linesize(2);
Y
for(y = 0; y < VIDEO_HEIGHT; y++)
{
for(x = 0; x < VIDEO_WIDTH; x++)
{
picture.data(0).put((y * pic_linesize_0 + x), data420[y * VIDEO_WIDTH + x]);
}
}
Cb and Cr
for(y = 0; y < VIDEO_HEIGHT / 2; y++) {
for(x = 0; x < VIDEO_WIDTH / 2; x++) {
picture.data(1).put((y * pic_linesize_1 + x), data420[data1_offset + y * VIDEO_WIDTH / 2 + x]);
picture.data(2).put((y * pic_linesize_2 + x), data420[data2_offset + y * VIDEO_WIDTH / 2 + x]);
}
}
(2)Encode
Encode the image into output_buffer
out_size = avcodec.avcodec_encode_video(mCodecCtx, new BytePointer(output_buffer), output_buffer_size, picture);
Log.d("TEST_VIDEO", "Encoded '" + out_size + "' bytes");
Delayed frames
for(; out_size > 0; i++) {
out_size = avcodec.avcodec_encode_video(mCodecCtx, new BytePointer(output_buffer), output_buffer_size, null);
Log.d("TEST_VIDEO", "Encoded '" + out_size + "' bytes");
fwrite(output_buffer, 1, out_size, file);
}
I’m still working on packaging the data, but the ongoing test project @ http://code.google.com/p/test-video-encode/