[FFmpeg-devel] [PATCH v2] avcodec/libwebpenc_anim: support setting the duration of the last frame
James Zern
jzern at google.com
Mon Mar 3 20:37:40 EET 2025
Hi,
On Fri, Feb 28, 2025 at 1:47 AM wangyaqiang via ffmpeg-devel
<ffmpeg-devel at ffmpeg.org> wrote:
>
> From: Wang Yaqiang <wangyaqiang03 at kuaishou.com>
>
> Signed-off-by: Wang Yaqiang <wangyaqiang03 at kuaishou.com>
> ---
> libavcodec/libwebpenc_animencoder.c | 17 +++++++++++++++--
> libavcodec/libwebpenc_common.c | 2 ++
> libavcodec/libwebpenc_common.h | 1 +
> 3 files changed, 18 insertions(+), 2 deletions(-)
>
> diff --git a/libavcodec/libwebpenc_animencoder.c b/libavcodec/libwebpenc_animencoder.c
> index c5361d7f92..558fcc246d 100644
> --- a/libavcodec/libwebpenc_animencoder.c
> +++ b/libavcodec/libwebpenc_animencoder.c
> @@ -38,6 +38,7 @@ typedef struct LibWebPAnimContext {
> WebPAnimEncoder *enc; // the main AnimEncoder object
> int64_t first_frame_pts; // pts of the first encoded frame.
> int64_t end_pts; // pts + duration of the last frame
> + int64_t prev_frame_pts; // pts of the previously encoded frame.
>
> void *first_frame_opaque;
> AVBufferRef *first_frame_opaque_ref;
> @@ -72,7 +73,19 @@ static int libwebp_anim_encode_frame(AVCodecContext *avctx, AVPacket *pkt,
> if (s->done) { // Second flush: return empty package to denote finish.
> *got_packet = 0;
> return 0;
> - } else { // First flush: assemble bitstream and return it.
> + } else {
> + if (s->cc.last_delay > 0) {
> + int timestamp_ms = avctx->time_base.num * (s->cc.last_delay + s->prev_frame_pts) * 1000 / avctx->time_base.den;
> + ret = WebPAnimEncoderAdd(s->enc, NULL, timestamp_ms, &s->cc.config);
> + if (!ret) {
> + av_log(avctx, AV_LOG_ERROR,
> + "Encoding WebP frame failed!\n");
> + ret = AVERROR_EXTERNAL;
> + goto end;
> + }
> + s->end_pts = s->prev_frame_pts + s->cc.last_delay;
> + }
> + // First flush: assemble bitstream and return it.
> WebPData assembled_data = { 0 };
> ret = WebPAnimEncoderAssemble(s->enc, &assembled_data);
> if (ret) {
> @@ -137,7 +150,7 @@ static int libwebp_anim_encode_frame(AVCodecContext *avctx, AVPacket *pkt,
>
> if (frame->pts != AV_NOPTS_VALUE)
> s->end_pts = frame->pts + frame->duration;
> -
> + s->prev_frame_pts = frame->pts;
> ret = 0;
> *got_packet = 0;
>
> diff --git a/libavcodec/libwebpenc_common.c b/libavcodec/libwebpenc_common.c
> index 80040ea9e3..13fb6aa690 100644
> --- a/libavcodec/libwebpenc_common.c
> +++ b/libavcodec/libwebpenc_common.c
> @@ -49,6 +49,8 @@ static const AVOption options[] = {
> { "cr_threshold","Conditional replenishment threshold", OFFSET(cr_threshold), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, INT_MAX, VE },
> { "cr_size" ,"Conditional replenishment block size", OFFSET(cr_size) , AV_OPT_TYPE_INT, { .i64 = 16 }, 0, 256, VE },
> { "quality" ,"Quality", OFFSET(quality), AV_OPT_TYPE_FLOAT, { .dbl = 75 }, 0, 100, VE },
> + { "final_delay", "Force delay (in centiseconds) after the last frame", OFFSET(last_delay),
> + AV_OPT_TYPE_INT, { .i64 = -1 }, -1, INT_MAX, VE },
This description doesn't sound right as the value looks like it's
being interpreted as if it's in the source timebase. Why not accept a
value in milliseconds and apply that to the final frame? There's also
a risk of overflow as timestamp_ms is an int, though that's ignored in
the existing calculation.
More information about the ffmpeg-devel
mailing list