diff mbox series

[FFmpeg-devel,v10,2/5] libavcodec/webp: add support for animated WebP

Message ID 20240205194142.37049-3-thilo.borgmann@mail.de
State New
Headers show
Series webp: add support for animated WebP decoding | expand

Checks

Context Check Description
yinshiyou/make_loongarch64 fail Make failed
andriy/make_x86 fail Make failed

Commit Message

Diego Felix de Souza via ffmpeg-devel Feb. 5, 2024, 7:41 p.m. UTC 
From: Josef Zlomek <josef@pex.com>

Fixes: 4907

Adds support for decoding of animated WebP.

The WebP decoder adds the animation related features according to the specs:
https://developers.google.com/speed/webp/docs/riff_container#animation
The frames of the animation may be smaller than the image canvas.
Therefore, the frame is decoded to a temporary frame,
then it is blended into the canvas, the canvas is copied to the output frame,
and finally the frame is disposed from the canvas.

The output to AV_PIX_FMT_YUVA420P/AV_PIX_FMT_YUV420P is still supported.
The background color is specified only as BGRA in the WebP file
so it is converted to YUVA if YUV formats are output.

Signed-off-by: Josef Zlomek <josef@pex.com>
---
 Changelog               |   1 +
 libavcodec/codec_desc.c |   3 +-
 libavcodec/version.h    |   2 +-
 libavcodec/webp.c       | 704 +++++++++++++++++++++++++++++++++++++---
 4 files changed, 654 insertions(+), 56 deletions(-)

Comments

Andreas Rheinhardt Feb. 19, 2024, 4:50 p.m. UTC | #1 
Thilo Borgmann via ffmpeg-devel:
> From: Josef Zlomek <josef@pex.com>
> 
> Fixes: 4907
> 
> Adds support for decoding of animated WebP.
> 
> The WebP decoder adds the animation related features according to the specs:
> https://developers.google.com/speed/webp/docs/riff_container#animation
> The frames of the animation may be smaller than the image canvas.
> Therefore, the frame is decoded to a temporary frame,
> then it is blended into the canvas, the canvas is copied to the output frame,
> and finally the frame is disposed from the canvas.
> 
> The output to AV_PIX_FMT_YUVA420P/AV_PIX_FMT_YUV420P is still supported.
> The background color is specified only as BGRA in the WebP file
> so it is converted to YUVA if YUV formats are output.
> 
> Signed-off-by: Josef Zlomek <josef@pex.com>
> ---
>  Changelog               |   1 +
>  libavcodec/codec_desc.c |   3 +-
>  libavcodec/version.h    |   2 +-
>  libavcodec/webp.c       | 704 +++++++++++++++++++++++++++++++++++++---
>  4 files changed, 654 insertions(+), 56 deletions(-)
> 

> +static int webp_decode_frame(AVCodecContext *avctx, AVFrame *p,
> +                             int *got_frame, AVPacket *avpkt)
> +{
> +    WebPContext *s = avctx->priv_data;
> +    AVFrame *canvas = s->canvas_frame.f;
> +    int ret;
> +    int key_frame = avpkt->flags & AV_PKT_FLAG_KEY;
> +
> +    *got_frame   = 0;
> +
> +    if (key_frame) {
> +        // The canvas is passed from one thread to another in a sequence
> +        // starting with a key frame followed by non-key frames.
> +        // The key frame reports progress 1,
> +        // the N-th non-key frame awaits progress N = s->await_progress
> +        // and reports progress N + 1.
> +        s->await_progress = 0;
> +    }
> +
> +    // reset the frame params
> +    s->anmf_flags = 0;
> +    s->width      = 0;
> +    s->height     = 0;
> +    s->pos_x      = 0;
> +    s->pos_y      = 0;
> +    s->has_alpha  = 0;
> +
> +    ret = webp_decode_frame_common(avctx, avpkt->data, avpkt->size, got_frame, key_frame);
> +    if (ret < 0)
> +        goto end;
> +
> +    if (s->vp8x_flags & VP8X_FLAG_ANIMATION) {
> +        // VP8 decoder might have changed the width and height of the frame
> +        AVFrame *frame  = s->frame;
> +        ret = av_frame_copy_props(canvas, frame);
> +        if (ret < 0)
> +            return ret;
> +
> +        ret = ff_set_dimensions(s->avctx, canvas->width, canvas->height);
> +        if (ret < 0)
> +            return ret;
> +
> +        s->avctx->pix_fmt = canvas->format;
> +    }
> +
> +    ff_thread_finish_setup(s->avctx);

1. Up until now, when decoding a series of stand-alone WebP pictures,
the multiple decoder instances don't wait for each other (because the
WebP decoder had no update_thread_context callback). You added such a
callback and now you are calling it after the main picture has already
been decoded, effectively serializing everything. You can test this for
yourself: Create lots of files via ffmpeg -i <input> -c:v libwebp -f
webp%d.webp and decode them (don't use -stream_loop on a single input
picture, as this will flush the decoder after every single picture, so
that everything is always serialized).
2. To fix this, ff_thread_finish_setup() needs to be called as soon as
possible. This means that you have to abandon the approach of letting
the inner VP8 decoder set the frame dimensions and then overwriting them
again in the WebP decoder.
3. Your WebP demuxer (from 4/5) splits an animation into a series of
packets. The problem is the design of the extended file format
(https://developers.google.com/speed/webp/docs/riff_container#extended_file_format):
Certain metadata chunks (in particular exif) that pertain to the whole
animation are stored only after the image data. If you split the input
into packets, the decoder won't be able to attach it to every frame
except the last one.
4. Due to this, splitting the input packet should be avoided. But this
causes great complications in the decoder: It now needs to output
multiple frames per packet (if said packet contained an animation). I
see three ways to do this:
a) Use the receive frame callback for this decoder. This will
necessitate changes to pthread_frame.c (which currently can't handle
receive_frame decoders) and even then it will have the further downside
that decoding a single animation is single-threaded when using
frame-threading (given that animations provide opportunity to decode
parts of a packet in parallel, they work naturally with slice threading,
not frame-threading). But non-animations should work fine as now.
b) Somehow implement this via AV_CODEC_CAP_OTHER_THREADS, potentially
via the AVExecutor API. This would have the upside that it could
dynamically switch between frame- and slice threading (depending upon
what the input is).
c) Accept packets containing whole animations, but use a BSF to split
the data so that the metadata arrives at the decoder before/together
with the real frame data. I am not sure whether this would necessitate
changes to the decode API, too (basically: if there are threads that are
currently not busy and the BSF has not signalled EAGAIN yet, then try to
extract another packet from the BSF).
Notice that the BSF I have in mind would not be a public BSF, but a
private one (given that the output of the BSF would be spec-incompliant
due to the wrong ordering it should not be public), i.e. one not
accessible via av_bsf_get_by_name() or av_bsf_iterate().

> +
> +    if (*got_frame) {
> +        if (!(s->vp8x_flags & VP8X_FLAG_ANIMATION)) {
> +            // no animation, output the decoded frame
> +            av_frame_move_ref(p, s->frame);
> +        } else {
> +            if (!key_frame) {
> +                ff_thread_await_progress(&s->canvas_frame, s->await_progress, 0);
> +
> +                ret = dispose_prev_frame_in_canvas(s);
> +                if (ret < 0)
> +                    goto end;
> +            }
> +
> +            ret = blend_frame_into_canvas(s);
> +            if (ret < 0)
> +                goto end;
> +
> +            ret = copy_canvas_to_frame(s, p, key_frame);
> +            if (ret < 0)
> +                goto end;
> +
> +            ff_thread_report_progress(&s->canvas_frame, s->await_progress + 1, 0);
> +        }
> +
> +        p->pts = avpkt->pts;
> +    }
> +
> +    ret = avpkt->size;
> +
> +end:
> +    av_frame_unref(s->frame);
> +    return ret;
>  }
>  
>  const FFCodec ff_webp_decoder = {
>      .p.name         = "webp",
>      CODEC_LONG_NAME("WebP image"),
>      .p.type         = AVMEDIA_TYPE_VIDEO,
>      .p.id           = AV_CODEC_ID_WEBP,
>      .priv_data_size = sizeof(WebPContext),
> +    UPDATE_THREAD_CONTEXT(webp_update_thread_context),
>      .init           = webp_decode_init,
>      FF_CODEC_DECODE_CB(webp_decode_frame),
>      .close          = webp_decode_close,
> +    .flush          = webp_decode_flush,
>      .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_FRAME_THREADS,
> -    .caps_internal  = FF_CODEC_CAP_ICC_PROFILES,
> +    .caps_internal  = FF_CODEC_CAP_ICC_PROFILES | FF_CODEC_CAP_ALLOCATE_PROGRESS,
>  };
James Almer Feb. 19, 2024, 5:07 p.m. UTC | #2 
On 2/19/2024 1:50 PM, Andreas Rheinhardt wrote:
> a) Use the receive frame callback for this decoder. This will
> necessitate changes to pthread_frame.c (which currently can't handle
> receive_frame decoders)

There's a patch for this that i think Anton wrote and then Timo 
improved. It was posted sometime ago.

> Notice that the BSF I have in mind would not be a public BSF, but a
> private one (given that the output of the BSF would be spec-incompliant
> due to the wrong ordering it should not be public), i.e. one not
> accessible via av_bsf_get_by_name() or av_bsf_iterate().

How do you propose to do this? And would you be willing to write the 
framework for it?
It certainly sounds useful, as existing bsfs like vp9_superframe_split 
already generate spec non-compliant packets, and they ideally should not 
be publicly available.
Andreas Rheinhardt Feb. 19, 2024, 6:01 p.m. UTC | #3 
James Almer:
> On 2/19/2024 1:50 PM, Andreas Rheinhardt wrote:
>> a) Use the receive frame callback for this decoder. This will
>> necessitate changes to pthread_frame.c (which currently can't handle
>> receive_frame decoders)
> 
> There's a patch for this that i think Anton wrote and then Timo
> improved. It was posted sometime ago.
> 
>> Notice that the BSF I have in mind would not be a public BSF, but a
>> private one (given that the output of the BSF would be spec-incompliant
>> due to the wrong ordering it should not be public), i.e. one not
>> accessible via av_bsf_get_by_name() or av_bsf_iterate().
> 
> How do you propose to do this? And would you be willing to write the
> framework for it?
> It certainly sounds useful, as existing bsfs like vp9_superframe_split
> already generate spec non-compliant packets, and they ideally should not
> be publicly available.

The way I envision is by using a const FFBitStreamFilter* instead of the
name of the bsf in FFCodec.bsfs. (Of course, one can keep the
open-by-string method (which also works for filterchains as long as they
only use static arguments), too; it is also possible (by currently
unneeded) to add bsfs during init with non-static arguments (based upon
codec options etc.).)
I'd be willing to write the code for this.

- Andreas
Thilo Borgmann Feb. 26, 2024, 3:54 p.m. UTC | #4 
Am 19.02.24 um 19:01 schrieb Andreas Rheinhardt:
> James Almer:
>> On 2/19/2024 1:50 PM, Andreas Rheinhardt wrote:
>>> a) Use the receive frame callback for this decoder. This will
>>> necessitate changes to pthread_frame.c (which currently can't handle
>>> receive_frame decoders)
>>
>> There's a patch for this that i think Anton wrote and then Timo
>> improved. It was posted sometime ago.
>>
>>> Notice that the BSF I have in mind would not be a public BSF, but a
>>> private one (given that the output of the BSF would be spec-incompliant
>>> due to the wrong ordering it should not be public), i.e. one not
>>> accessible via av_bsf_get_by_name() or av_bsf_iterate().
>>
>> How do you propose to do this? And would you be willing to write the
>> framework for it?
>> It certainly sounds useful, as existing bsfs like vp9_superframe_split
>> already generate spec non-compliant packets, and they ideally should not
>> be publicly available.
> 
> The way I envision is by using a const FFBitStreamFilter* instead of the
> name of the bsf in FFCodec.bsfs. (Of course, one can keep the
> open-by-string method (which also works for filterchains as long as they
> only use static arguments), too; it is also possible (by currently
> unneeded) to add bsfs during init with non-static arguments (based upon
> codec options etc.).)
> I'd be willing to write the code for this.

If that is useful for other codecs as well, let's do that.
Generating one packet per animation containing all frames+meta should be easy enough in the demuxer.

What do you expect the bsf to do exactly?
IIUC generate packets with one frame + meta for all frames in the animation from the big packet coming from the demuxer?

Thanks,
Thilo
Thilo Borgmann Feb. 26, 2024, 3:54 p.m. UTC | #5 
Hi,

Am 19.02.24 um 17:50 schrieb Andreas Rheinhardt:
> Thilo Borgmann via ffmpeg-devel:
>> From: Josef Zlomek <josef@pex.com>
>>
>> Fixes: 4907
>>
>> Adds support for decoding of animated WebP.
>>
>> The WebP decoder adds the animation related features according to the specs:
>> https://developers.google.com/speed/webp/docs/riff_container#animation
>> The frames of the animation may be smaller than the image canvas.
>> Therefore, the frame is decoded to a temporary frame,
>> then it is blended into the canvas, the canvas is copied to the output frame,
>> and finally the frame is disposed from the canvas.
>>
>> The output to AV_PIX_FMT_YUVA420P/AV_PIX_FMT_YUV420P is still supported.
>> The background color is specified only as BGRA in the WebP file
>> so it is converted to YUVA if YUV formats are output.
>>
>> Signed-off-by: Josef Zlomek <josef@pex.com>
>> ---
>>   Changelog               |   1 +
>>   libavcodec/codec_desc.c |   3 +-
>>   libavcodec/version.h    |   2 +-
>>   libavcodec/webp.c       | 704 +++++++++++++++++++++++++++++++++++++---
>>   4 files changed, 654 insertions(+), 56 deletions(-)
>>
> 
>> +static int webp_decode_frame(AVCodecContext *avctx, AVFrame *p,
>> +                             int *got_frame, AVPacket *avpkt)
>> +{
>> +    WebPContext *s = avctx->priv_data;
>> +    AVFrame *canvas = s->canvas_frame.f;
>> +    int ret;
>> +    int key_frame = avpkt->flags & AV_PKT_FLAG_KEY;
>> +
>> +    *got_frame   = 0;
>> +
>> +    if (key_frame) {
>> +        // The canvas is passed from one thread to another in a sequence
>> +        // starting with a key frame followed by non-key frames.
>> +        // The key frame reports progress 1,
>> +        // the N-th non-key frame awaits progress N = s->await_progress
>> +        // and reports progress N + 1.
>> +        s->await_progress = 0;
>> +    }
>> +
>> +    // reset the frame params
>> +    s->anmf_flags = 0;
>> +    s->width      = 0;
>> +    s->height     = 0;
>> +    s->pos_x      = 0;
>> +    s->pos_y      = 0;
>> +    s->has_alpha  = 0;
>> +
>> +    ret = webp_decode_frame_common(avctx, avpkt->data, avpkt->size, got_frame, key_frame);
>> +    if (ret < 0)
>> +        goto end;
>> +
>> +    if (s->vp8x_flags & VP8X_FLAG_ANIMATION) {
>> +        // VP8 decoder might have changed the width and height of the frame
>> +        AVFrame *frame  = s->frame;
>> +        ret = av_frame_copy_props(canvas, frame);
>> +        if (ret < 0)
>> +            return ret;
>> +
>> +        ret = ff_set_dimensions(s->avctx, canvas->width, canvas->height);
>> +        if (ret < 0)
>> +            return ret;
>> +
>> +        s->avctx->pix_fmt = canvas->format;
>> +    }
>> +
>> +    ff_thread_finish_setup(s->avctx);
> 

> 1. Up until now, when decoding a series of stand-alone WebP pictures,
> the multiple decoder instances don't wait for each other (because the
> WebP decoder had no update_thread_context callback). You added such a
> callback and now you are calling it after the main picture has already
> been decoded, effectively serializing everything. You can test this for
> yourself: Create lots of files via ffmpeg -i <input> -c:v libwebp -f
> webp%d.webp and decode them (don't use -stream_loop on a single input
> picture, as this will flush the decoder after every single picture, so
> that everything is always serialized).
> 2. To fix this, ff_thread_finish_setup() needs to be called as soon as
> possible. This means that you have to abandon the approach of letting
> the inner VP8 decoder set the frame dimensions and then overwriting them
> again in the WebP decoder.

so I tried to move the ff_thread_finish_setup() call and avoid updating the AVCodecContext after decoding.

With the integrated decoder like here, ff_vp8_decode_frame() writes into the AVCodecContext while decoding, so ff_thread_finish_setup() can only be called afterwards -> quasi sequentially. I do see decoding speed for many files at once dropping from 100x to 16x compared to master.

With the decoder decoupled, I can actually move ff_thread_finish_setup() before send_packet() (in contrast to v9) -> next thread can start before decoding the frame.
I don't see a penalty with that way on decoding many files at once compared to master, as expected.

Is that reason enough to actually decouple the vp8 decoder?


Thanks,
Thilo
diff mbox series

Patch

diff --git a/Changelog b/Changelog
index c5fb21d198..87f4dbdef4 100644
--- a/Changelog
+++ b/Changelog
@@ -62,6 +62,7 @@  version 6.1:
 - ffprobe XML output schema changed to account for multiple
   variable-fields elements within the same parent element
 - ffprobe -output_format option added as an alias of -of
+- animated WebP decoder
 
 
 version 6.0:
diff --git a/libavcodec/codec_desc.c b/libavcodec/codec_desc.c
index 033344304c..0f72769093 100644
--- a/libavcodec/codec_desc.c
+++ b/libavcodec/codec_desc.c
@@ -1259,8 +1259,7 @@  static const AVCodecDescriptor codec_descriptors[] = {
         .type      = AVMEDIA_TYPE_VIDEO,
         .name      = "webp",
         .long_name = NULL_IF_CONFIG_SMALL("WebP"),
-        .props     = AV_CODEC_PROP_INTRA_ONLY | AV_CODEC_PROP_LOSSY |
-                     AV_CODEC_PROP_LOSSLESS,
+        .props     = AV_CODEC_PROP_LOSSY | AV_CODEC_PROP_LOSSLESS,
         .mime_types= MT("image/webp"),
     },
     {
diff --git a/libavcodec/version.h b/libavcodec/version.h
index f2f14eaed1..ecdbc51c74 100644
--- a/libavcodec/version.h
+++ b/libavcodec/version.h
@@ -30,7 +30,7 @@ 
 #include "version_major.h"
 
 #define LIBAVCODEC_VERSION_MINOR  39
-#define LIBAVCODEC_VERSION_MICRO 100
+#define LIBAVCODEC_VERSION_MICRO 101
 
 #define LIBAVCODEC_VERSION_INT  AV_VERSION_INT(LIBAVCODEC_VERSION_MAJOR, \
                                                LIBAVCODEC_VERSION_MINOR, \
diff --git a/libavcodec/webp.c b/libavcodec/webp.c
index 4fd107aa0c..4119ae679d 100644
--- a/libavcodec/webp.c
+++ b/libavcodec/webp.c
@@ -35,12 +35,15 @@ 
  * Exif metadata
  * ICC profile
  *
+ * @author Josef Zlomek, Pexeso Inc. <josef@pex.com>
+ * Animation
+ *
  * Unimplemented:
- *   - Animation
  *   - XMP metadata
  */
 
 #include "libavutil/imgutils.h"
+#include "libavutil/colorspace.h"
 
 #define BITSTREAM_READER_LE
 #include "avcodec.h"
@@ -67,6 +70,14 @@ 
 #define NUM_SHORT_DISTANCES             120
 #define MAX_HUFFMAN_CODE_LENGTH         15
 
+#define ANMF_DISPOSAL_METHOD            0x01
+#define ANMF_DISPOSAL_METHOD_UNCHANGED  0x00
+#define ANMF_DISPOSAL_METHOD_BACKGROUND 0x01
+
+#define ANMF_BLENDING_METHOD            0x02
+#define ANMF_BLENDING_METHOD_ALPHA      0x00
+#define ANMF_BLENDING_METHOD_OVERWRITE  0x02
+
 static const uint16_t alphabet_sizes[HUFFMAN_CODES_PER_META_CODE] = {
     NUM_LITERAL_CODES + NUM_LENGTH_CODES,
     NUM_LITERAL_CODES, NUM_LITERAL_CODES, NUM_LITERAL_CODES,
@@ -191,6 +202,8 @@  typedef struct ImageContext {
 typedef struct WebPContext {
     VP8Context v;                       /* VP8 Context used for lossy decoding */
     GetBitContext gb;                   /* bitstream reader for main image chunk */
+    ThreadFrame canvas_frame;           /* ThreadFrame for canvas */
+    AVFrame *frame;                     /* AVFrame for decoded frame */
     AVFrame *alpha_frame;               /* AVFrame for alpha data decompressed from VP8L */
     AVPacket *pkt;                      /* AVPacket to be passed to the underlying VP8 decoder */
     AVCodecContext *avctx;              /* parent AVCodecContext */
@@ -204,7 +217,22 @@  typedef struct WebPContext {
     int has_iccp;                       /* set after an ICCP chunk has been processed */
     int width;                          /* image width */
     int height;                         /* image height */
-    int lossless;                       /* indicates lossless or lossy */
+    int vp8x_flags;                     /* global flags from VP8X chunk */
+    int canvas_width;                   /* canvas width */
+    int canvas_height;                  /* canvas height */
+    int anmf_flags;                     /* frame flags from ANMF chunk */
+    int pos_x;                          /* frame position X */
+    int pos_y;                          /* frame position Y */
+    int prev_anmf_flags;                /* previous frame flags from ANMF chunk */
+    int prev_width;                     /* previous frame width */
+    int prev_height;                    /* previous frame height */
+    int prev_pos_x;                     /* previous frame position X */
+    int prev_pos_y;                     /* previous frame position Y */
+    int await_progress;                 /* value of progress to wait for */
+    uint8_t background_argb[4];         /* background color in ARGB format */
+    uint8_t background_yuva[4];         /* background color in YUVA format */
+    const uint8_t *background_data[4];  /* "planes" for background color in YUVA format */
+    uint8_t transparent_yuva[4];        /* transparent black in YUVA format */
 
     int nb_transforms;                  /* number of transforms */
     enum TransformType transforms[4];   /* transformations used in the image, in order */
@@ -1089,7 +1117,6 @@  static int vp8_lossless_decode_frame(AVCodecContext *avctx, AVFrame *p,
     int w, h, ret, i, used;
 
     if (!is_alpha_chunk) {
-        s->lossless = 1;
         avctx->pix_fmt = AV_PIX_FMT_ARGB;
     }
 
@@ -1303,7 +1330,6 @@  static int vp8_lossy_decode_frame(AVCodecContext *avctx, AVFrame *p,
         s->v.actually_webp = 1;
     }
     avctx->pix_fmt = s->has_alpha ? AV_PIX_FMT_YUVA420P : AV_PIX_FMT_YUV420P;
-    s->lossless = 0;
 
     if (data_size > INT_MAX) {
         av_log(avctx, AV_LOG_ERROR, "unsupported chunk size\n");
@@ -1332,40 +1358,17 @@  static int vp8_lossy_decode_frame(AVCodecContext *avctx, AVFrame *p,
     return ret;
 }
 
-static int webp_decode_frame(AVCodecContext *avctx, AVFrame *p,
-                             int *got_frame, AVPacket *avpkt)
+int init_canvas_frame(WebPContext *s, int format, int key_frame);
+
+static int webp_decode_frame_common(AVCodecContext *avctx, uint8_t *data, int size,
+                                    int *got_frame, int key_frame)
 {
     WebPContext *s = avctx->priv_data;
     GetByteContext gb;
     int ret;
     uint32_t chunk_type, chunk_size;
-    int vp8x_flags = 0;
 
-    s->avctx     = avctx;
-    s->width     = 0;
-    s->height    = 0;
-    *got_frame   = 0;
-    s->has_alpha = 0;
-    s->has_exif  = 0;
-    s->has_iccp  = 0;
-    bytestream2_init(&gb, avpkt->data, avpkt->size);
-
-    if (bytestream2_get_bytes_left(&gb) < 12)
-        return AVERROR_INVALIDDATA;
-
-    if (bytestream2_get_le32(&gb) != MKTAG('R', 'I', 'F', 'F')) {
-        av_log(avctx, AV_LOG_ERROR, "missing RIFF tag\n");
-        return AVERROR_INVALIDDATA;
-    }
-
-    chunk_size = bytestream2_get_le32(&gb);
-    if (bytestream2_get_bytes_left(&gb) < chunk_size)
-        return AVERROR_INVALIDDATA;
-
-    if (bytestream2_get_le32(&gb) != MKTAG('W', 'E', 'B', 'P')) {
-        av_log(avctx, AV_LOG_ERROR, "missing WEBP tag\n");
-        return AVERROR_INVALIDDATA;
-    }
+    bytestream2_init(&gb, data, size);
 
     while (bytestream2_get_bytes_left(&gb) > 8) {
         char chunk_str[5] = { 0 };
@@ -1376,6 +1379,10 @@  static int webp_decode_frame(AVCodecContext *avctx, AVFrame *p,
             return AVERROR_INVALIDDATA;
         chunk_size += chunk_size & 1;
 
+        // we need to dive into RIFF chunk
+        if (chunk_type == MKTAG('R', 'I', 'F', 'F'))
+            chunk_size = 4;
+
         if (bytestream2_get_bytes_left(&gb) < chunk_size) {
            /* we seem to be running out of data, but it could also be that the
               bitstream has trailing junk leading to bogus chunk_size. */
@@ -1383,10 +1390,26 @@  static int webp_decode_frame(AVCodecContext *avctx, AVFrame *p,
         }
 
         switch (chunk_type) {
+        case MKTAG('R', 'I', 'F', 'F'):
+            if (bytestream2_get_le32(&gb) != MKTAG('W', 'E', 'B', 'P')) {
+                av_log(avctx, AV_LOG_ERROR, "missing WEBP tag\n");
+                return AVERROR_INVALIDDATA;
+            }
+            s->vp8x_flags    = 0;
+            s->canvas_width  = 0;
+            s->canvas_height = 0;
+            s->has_exif      = 0;
+            s->has_iccp      = 0;
+            ff_thread_release_ext_buffer(&s->canvas_frame);
+            break;
         case MKTAG('V', 'P', '8', ' '):
             if (!*got_frame) {
-                ret = vp8_lossy_decode_frame(avctx, p, got_frame,
-                                             avpkt->data + bytestream2_tell(&gb),
+                ret = init_canvas_frame(s, AV_PIX_FMT_YUVA420P, key_frame);
+                if (ret < 0)
+                    return ret;
+
+                ret = vp8_lossy_decode_frame(avctx, s->frame, got_frame,
+                                             data + bytestream2_tell(&gb),
                                              chunk_size);
                 if (ret < 0)
                     return ret;
@@ -1395,8 +1418,12 @@  static int webp_decode_frame(AVCodecContext *avctx, AVFrame *p,
             break;
         case MKTAG('V', 'P', '8', 'L'):
             if (!*got_frame) {
-                ret = vp8_lossless_decode_frame(avctx, p, got_frame,
-                                                avpkt->data + bytestream2_tell(&gb),
+                ret = init_canvas_frame(s, AV_PIX_FMT_ARGB, key_frame);
+                if (ret < 0)
+                    return ret;
+
+                ret = vp8_lossless_decode_frame(avctx, s->frame, got_frame,
+                                                data + bytestream2_tell(&gb),
                                                 chunk_size, 0);
                 if (ret < 0)
                     return ret;
@@ -1405,14 +1432,16 @@  static int webp_decode_frame(AVCodecContext *avctx, AVFrame *p,
             bytestream2_skip(&gb, chunk_size);
             break;
         case MKTAG('V', 'P', '8', 'X'):
-            if (s->width || s->height || *got_frame) {
+            if (s->canvas_width || s->canvas_height || *got_frame) {
                 av_log(avctx, AV_LOG_ERROR, "Canvas dimensions are already set\n");
                 return AVERROR_INVALIDDATA;
             }
-            vp8x_flags = bytestream2_get_byte(&gb);
+            s->vp8x_flags = bytestream2_get_byte(&gb);
             bytestream2_skip(&gb, 3);
             s->width  = bytestream2_get_le24(&gb) + 1;
             s->height = bytestream2_get_le24(&gb) + 1;
+            s->canvas_width  = s->width;
+            s->canvas_height = s->height;
             ret = av_image_check_size(s->width, s->height, 0, avctx);
             if (ret < 0)
                 return ret;
@@ -1420,7 +1449,7 @@  static int webp_decode_frame(AVCodecContext *avctx, AVFrame *p,
         case MKTAG('A', 'L', 'P', 'H'): {
             int alpha_header, filter_m, compression;
 
-            if (!(vp8x_flags & VP8X_FLAG_ALPHA)) {
+            if (!(s->vp8x_flags & VP8X_FLAG_ALPHA)) {
                 av_log(avctx, AV_LOG_WARNING,
                        "ALPHA chunk present, but alpha bit not set in the "
                        "VP8X header\n");
@@ -1429,8 +1458,9 @@  static int webp_decode_frame(AVCodecContext *avctx, AVFrame *p,
                 av_log(avctx, AV_LOG_ERROR, "invalid ALPHA chunk size\n");
                 return AVERROR_INVALIDDATA;
             }
+
             alpha_header       = bytestream2_get_byte(&gb);
-            s->alpha_data      = avpkt->data + bytestream2_tell(&gb);
+            s->alpha_data      = data + bytestream2_tell(&gb);
             s->alpha_data_size = chunk_size - 1;
             bytestream2_skip(&gb, s->alpha_data_size);
 
@@ -1457,14 +1487,13 @@  static int webp_decode_frame(AVCodecContext *avctx, AVFrame *p,
                 av_log(avctx, AV_LOG_VERBOSE, "Ignoring extra EXIF chunk\n");
                 goto exif_end;
             }
-            if (!(vp8x_flags & VP8X_FLAG_EXIF_METADATA))
+            if (!(s->vp8x_flags & VP8X_FLAG_EXIF_METADATA))
                 av_log(avctx, AV_LOG_WARNING,
                        "EXIF chunk present, but Exif bit not set in the "
                        "VP8X header\n");
 
             s->has_exif = 1;
-            bytestream2_init(&exif_gb, avpkt->data + exif_offset,
-                             avpkt->size - exif_offset);
+            bytestream2_init(&exif_gb, data + exif_offset, size - exif_offset);
             if (ff_tdecode_header(&exif_gb, &le, &ifd_offset) < 0) {
                 av_log(avctx, AV_LOG_ERROR, "invalid TIFF header "
                        "in Exif data\n");
@@ -1477,7 +1506,7 @@  static int webp_decode_frame(AVCodecContext *avctx, AVFrame *p,
                 goto exif_end;
             }
 
-            av_dict_copy(&p->metadata, exif_metadata, 0);
+            av_dict_copy(&s->frame->metadata, exif_metadata, 0);
 
 exif_end:
             av_dict_free(&exif_metadata);
@@ -1492,21 +1521,64 @@  exif_end:
                 bytestream2_skip(&gb, chunk_size);
                 break;
             }
-            if (!(vp8x_flags & VP8X_FLAG_ICC))
+            if (!(s->vp8x_flags & VP8X_FLAG_ICC))
                 av_log(avctx, AV_LOG_WARNING,
                        "ICCP chunk present, but ICC Profile bit not set in the "
                        "VP8X header\n");
 
             s->has_iccp = 1;
-            sd = av_frame_new_side_data(p, AV_FRAME_DATA_ICC_PROFILE, chunk_size);
+            sd = av_frame_new_side_data(s->frame, AV_FRAME_DATA_ICC_PROFILE, chunk_size);
             if (!sd)
                 return AVERROR(ENOMEM);
 
             bytestream2_get_buffer(&gb, sd->data, chunk_size);
             break;
         }
-        case MKTAG('A', 'N', 'I', 'M'):
+        case MKTAG('A', 'N', 'I', 'M'): {
+            const AVPixFmtDescriptor *desc;
+            int a, r, g, b;
+            if (!(s->vp8x_flags & VP8X_FLAG_ANIMATION)) {
+                av_log(avctx, AV_LOG_WARNING,
+                       "ANIM chunk present, but animation bit not set in the "
+                       "VP8X header\n");
+            }
+            // background is stored as BGRA, we need ARGB
+            s->background_argb[3] = b = bytestream2_get_byte(&gb);
+            s->background_argb[2] = g = bytestream2_get_byte(&gb);
+            s->background_argb[1] = r = bytestream2_get_byte(&gb);
+            s->background_argb[0] = a = bytestream2_get_byte(&gb);
+
+            // convert the background color to YUVA
+            desc = av_pix_fmt_desc_get(AV_PIX_FMT_YUVA420P);
+            s->background_yuva[desc->comp[0].plane] = RGB_TO_Y_CCIR(r, g, b);
+            s->background_yuva[desc->comp[1].plane] = RGB_TO_U_CCIR(r, g, b, 0);
+            s->background_yuva[desc->comp[2].plane] = RGB_TO_V_CCIR(r, g, b, 0);
+            s->background_yuva[desc->comp[3].plane] = a;
+
+            bytestream2_skip(&gb, 2); // loop count is ignored
+            break;
+        }
         case MKTAG('A', 'N', 'M', 'F'):
+            if (!(s->vp8x_flags & VP8X_FLAG_ANIMATION)) {
+                av_log(avctx, AV_LOG_WARNING,
+                       "ANMF chunk present, but animation bit not set in the "
+                       "VP8X header\n");
+            }
+            s->pos_x      = bytestream2_get_le24(&gb) * 2;
+            s->pos_y      = bytestream2_get_le24(&gb) * 2;
+            s->width      = bytestream2_get_le24(&gb) + 1;
+            s->height     = bytestream2_get_le24(&gb) + 1;
+            bytestream2_skip(&gb, 3);   // duration
+            s->anmf_flags = bytestream2_get_byte(&gb);
+
+            if (s->width  + s->pos_x > s->canvas_width ||
+                s->height + s->pos_y > s->canvas_height) {
+                av_log(avctx, AV_LOG_ERROR,
+                       "frame does not fit into canvas\n");
+                return AVERROR_INVALIDDATA;
+            }
+            s->vp8x_flags |= VP8X_FLAG_ANIMATION;
+            break;
         case MKTAG('X', 'M', 'P', ' '):
             AV_WL32(chunk_str, chunk_type);
             av_log(avctx, AV_LOG_WARNING, "skipping unsupported chunk: %s\n",
@@ -1522,21 +1594,503 @@  exif_end:
         }
     }
 
-    if (!*got_frame) {
-        av_log(avctx, AV_LOG_ERROR, "image data not found\n");
-        return AVERROR_INVALIDDATA;
+    return size;
+}
+
+int init_canvas_frame(WebPContext *s, int format, int key_frame)
+{
+    AVFrame *canvas = s->canvas_frame.f;
+    int height;
+    int ret;
+
+    // canvas is needed only for animation
+    if (!(s->vp8x_flags & VP8X_FLAG_ANIMATION))
+        return 0;
+
+    // avoid init for non-key frames whose format and size did not change
+    if (!key_frame &&
+        canvas->data[0] &&
+        canvas->format == format &&
+        canvas->width  == s->canvas_width &&
+        canvas->height == s->canvas_height)
+        return 0;
+
+    // canvas changes within IPPP sequences will loose thread sync
+    // because of the ThreadFrame reallocation and will wait forever
+    // so if frame-threading is used, forbid canvas changes and unlock
+    // previous frames
+    if (!key_frame && canvas->data[0]) {
+        if (s->avctx->thread_count > 1) {
+            av_log(s->avctx, AV_LOG_WARNING, "Canvas change detected. The output will be damaged. Use -threads 1 to try decoding with best effort.\n");
+            // unlock previous frames that have sent an _await() call
+            ff_thread_report_progress(&s->canvas_frame, INT_MAX, 0);
+            return AVERROR_PATCHWELCOME;
+        } else {
+            // warn for damaged frames
+            av_log(s->avctx, AV_LOG_WARNING, "Canvas change detected. The output will be damaged.\n");
+        }
+    }
+
+    s->avctx->pix_fmt = format;
+    canvas->format    = format;
+    canvas->width     = s->canvas_width;
+    canvas->height    = s->canvas_height;
+
+    // VP8 decoder changed the width and height in AVCodecContext.
+    // Change it back to the canvas size.
+    ret = ff_set_dimensions(s->avctx, s->canvas_width, s->canvas_height);
+    if (ret < 0)
+        return ret;
+
+    ff_thread_release_ext_buffer(s->avctx, &s->canvas_frame);
+    ret = ff_thread_get_ext_buffer(s->avctx, &s->canvas_frame, AV_GET_BUFFER_FLAG_REF);
+    if (ret < 0)
+        return ret;
+
+    if (canvas->format == AV_PIX_FMT_ARGB) {
+        height = canvas->height;
+        memset(canvas->data[0], 0, height * canvas->linesize[0]);
+    } else /* if (canvas->format == AV_PIX_FMT_YUVA420P) */ {
+        const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(canvas->format);
+        for (int comp = 0; comp < desc->nb_components; comp++) {
+            int plane = desc->comp[comp].plane;
+
+            if (comp == 1 || comp == 2)
+                height = AV_CEIL_RSHIFT(canvas->height, desc->log2_chroma_h);
+            else
+                height = FFALIGN(canvas->height, 1 << desc->log2_chroma_h);
+
+            memset(canvas->data[plane], s->transparent_yuva[plane],
+                   height * canvas->linesize[plane]);
+        }
+    }
+
+    return 0;
+}
+
+/*
+ * Blend src1 (foreground) and src2 (background) into dest, in ARGB format.
+ * width, height are the dimensions of src1
+ * pos_x, pos_y is the position in src2 and in dest
+ */
+static void blend_alpha_argb(uint8_t *dest_data[4], int dest_linesize[4],
+                             const uint8_t *src1_data[4], int src1_linesize[4],
+                             const uint8_t *src2_data[4], int src2_linesize[4],
+                             int src2_step[4],
+                             int width, int height, int pos_x, int pos_y)
+{
+    for (int y = 0; y < height; y++) {
+        const uint8_t *src1 = src1_data[0] + y * src1_linesize[0];
+        const uint8_t *src2 = src2_data[0] + (y + pos_y) * src2_linesize[0] + pos_x * src2_step[0];
+        uint8_t       *dest = dest_data[0] + (y + pos_y) * dest_linesize[0] + pos_x * sizeof(uint32_t);
+        for (int x = 0; x < width; x++) {
+            int src1_alpha = src1[0];
+            int src2_alpha = src2[0];
+
+            if (src1_alpha == 255) {
+                memcpy(dest, src1, sizeof(uint32_t));
+            } else if (src1_alpha + src2_alpha == 0) {
+                memset(dest, 0, sizeof(uint32_t));
+            } else {
+                int tmp_alpha = src2_alpha - ROUNDED_DIV(src1_alpha * src2_alpha, 255);
+                int blend_alpha = src1_alpha + tmp_alpha;
+
+                dest[0] = blend_alpha;
+                dest[1] = ROUNDED_DIV(src1[1] * src1_alpha + src2[1] * tmp_alpha, blend_alpha);
+                dest[2] = ROUNDED_DIV(src1[2] * src1_alpha + src2[2] * tmp_alpha, blend_alpha);
+                dest[3] = ROUNDED_DIV(src1[3] * src1_alpha + src2[3] * tmp_alpha, blend_alpha);
+            }
+            src1 += sizeof(uint32_t);
+            src2 += src2_step[0];
+            dest += sizeof(uint32_t);
+        }
+    }
+}
+
+/*
+ * Blend src1 (foreground) and src2 (background) into dest, in YUVA format.
+ * width, height are the dimensions of src1
+ * pos_x, pos_y is the position in src2 and in dest
+ */
+static void blend_alpha_yuva(WebPContext *s,
+                             uint8_t *dest_data[4], int dest_linesize[4],
+                             const uint8_t *src1_data[4], int src1_linesize[4],
+                             int src1_format,
+                             const uint8_t *src2_data[4], int src2_linesize[4],
+                             int src2_step[4],
+                             int width, int height, int pos_x, int pos_y)
+{
+    const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(src1_format);
+
+    int plane_y = desc->comp[0].plane;
+    int plane_u = desc->comp[1].plane;
+    int plane_v = desc->comp[2].plane;
+    int plane_a = desc->comp[3].plane;
+
+    // blend U & V planes first, because the later step may modify alpha plane
+    int w  = AV_CEIL_RSHIFT(width,  desc->log2_chroma_w);
+    int h  = AV_CEIL_RSHIFT(height, desc->log2_chroma_h);
+    int px = AV_CEIL_RSHIFT(pos_x,  desc->log2_chroma_w);
+    int py = AV_CEIL_RSHIFT(pos_y,  desc->log2_chroma_h);
+    int tile_w = 1 << desc->log2_chroma_w;
+    int tile_h = 1 << desc->log2_chroma_h;
+
+    for (int y = 0; y < h; y++) {
+        const uint8_t *src1_u = src1_data[plane_u] + y * src1_linesize[plane_u];
+        const uint8_t *src1_v = src1_data[plane_v] + y * src1_linesize[plane_v];
+        const uint8_t *src2_u = src2_data[plane_u] + (y + py) * src2_linesize[plane_u] + px * src2_step[plane_u];
+        const uint8_t *src2_v = src2_data[plane_v] + (y + py) * src2_linesize[plane_v] + px * src2_step[plane_v];
+        uint8_t       *dest_u = dest_data[plane_u] + (y + py) * dest_linesize[plane_u] + px;
+        uint8_t       *dest_v = dest_data[plane_v] + (y + py) * dest_linesize[plane_v] + px;
+        for (int x = 0; x < w; x++) {
+            // calculate the average alpha of the tile
+            int src1_alpha = 0;
+            int src2_alpha = 0;
+            for (int yy = 0; yy < tile_h; yy++) {
+                for (int xx = 0; xx < tile_w; xx++) {
+                    src1_alpha += src1_data[plane_a][(y * tile_h + yy) * src1_linesize[plane_a] +
+                                                     (x * tile_w + xx)];
+                    src2_alpha += src2_data[plane_a][((y + py) * tile_h + yy) * src2_linesize[plane_a] +
+                                                     ((x + px) * tile_w + xx) * src2_step[plane_a]];
+                }
+            }
+            src1_alpha = AV_CEIL_RSHIFT(src1_alpha, desc->log2_chroma_w + desc->log2_chroma_h);
+            src2_alpha = AV_CEIL_RSHIFT(src2_alpha, desc->log2_chroma_w + desc->log2_chroma_h);
+
+            if (src1_alpha == 255) {
+                *dest_u = *src1_u;
+                *dest_v = *src1_v;
+            } else if (src1_alpha + src2_alpha == 0) {
+                *dest_u = s->transparent_yuva[plane_u];
+                *dest_v = s->transparent_yuva[plane_v];
+            } else {
+                int tmp_alpha = src2_alpha - ROUNDED_DIV(src1_alpha * src2_alpha, 255);
+                int blend_alpha = src1_alpha + tmp_alpha;
+                *dest_u = ROUNDED_DIV(*src1_u * src1_alpha + *src2_u * tmp_alpha, blend_alpha);
+                *dest_v = ROUNDED_DIV(*src1_v * src1_alpha + *src2_v * tmp_alpha, blend_alpha);
+            }
+            src1_u++;
+            src1_v++;
+            src2_u += src2_step[plane_u];
+            src2_v += src2_step[plane_v];
+            dest_u++;
+            dest_v++;
+        }
+    }
+
+    // blend Y & A planes
+    for (int y = 0; y < height; y++) {
+        const uint8_t *src1_y = src1_data[plane_y] + y * src1_linesize[plane_y];
+        const uint8_t *src1_a = src1_data[plane_a] + y * src1_linesize[plane_a];
+        const uint8_t *src2_y = src2_data[plane_y] + (y + pos_y) * src2_linesize[plane_y] + pos_x * src2_step[plane_y];
+        const uint8_t *src2_a = src2_data[plane_a] + (y + pos_y) * src2_linesize[plane_a] + pos_x * src2_step[plane_a];
+        uint8_t       *dest_y = dest_data[plane_y] + (y + pos_y) * dest_linesize[plane_y] + pos_x;
+        uint8_t       *dest_a = dest_data[plane_a] + (y + pos_y) * dest_linesize[plane_a] + pos_x;
+        for (int x = 0; x < width; x++) {
+            int src1_alpha = *src1_a;
+            int src2_alpha = *src2_a;
+
+            if (src1_alpha == 255) {
+                *dest_y = *src1_y;
+                *dest_a = 255;
+            } else if (src1_alpha + src2_alpha == 0) {
+                *dest_y = s->transparent_yuva[plane_y];
+                *dest_a = 0;
+            } else {
+                int tmp_alpha = src2_alpha - ROUNDED_DIV(src1_alpha * src2_alpha, 255);
+                int blend_alpha = src1_alpha + tmp_alpha;
+                *dest_y = ROUNDED_DIV(*src1_y * src1_alpha + *src2_y * tmp_alpha, blend_alpha);
+                *dest_a = blend_alpha;
+            }
+            src1_y++;
+            src1_a++;
+            src2_y += src2_step[plane_y];
+            src2_a += src2_step[plane_a];
+            dest_y++;
+            dest_a++;
+        }
+    }
+}
+
+static int blend_frame_into_canvas(WebPContext *s)
+{
+    AVFrame *canvas = s->canvas_frame.f;
+    AVFrame *frame  = s->frame;
+    int width, height;
+    int pos_x, pos_y;
+
+    if ((s->anmf_flags & ANMF_BLENDING_METHOD) == ANMF_BLENDING_METHOD_OVERWRITE
+        || frame->format == AV_PIX_FMT_YUV420P) {
+        // do not blend, overwrite
+
+        if (canvas->format == AV_PIX_FMT_ARGB) {
+            width  = s->width;
+            height = s->height;
+            pos_x  = s->pos_x;
+            pos_y  = s->pos_y;
+
+            for (int y = 0; y < height; y++) {
+                const uint32_t *src = (uint32_t *) (frame->data[0] + y * frame->linesize[0]);
+                uint32_t *dst = (uint32_t *) (canvas->data[0] + (y + pos_y) * canvas->linesize[0]) + pos_x;
+                memcpy(dst, src, width * sizeof(uint32_t));
+            }
+        } else /* if (canvas->format == AV_PIX_FMT_YUVA420P) */ {
+            const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(frame->format);
+            int plane;
+
+            for (int comp = 0; comp < desc->nb_components; comp++) {
+                plane  = desc->comp[comp].plane;
+                width  = s->width;
+                height = s->height;
+                pos_x  = s->pos_x;
+                pos_y  = s->pos_y;
+                if (comp == 1 || comp == 2) {
+                    width  = AV_CEIL_RSHIFT(width,  desc->log2_chroma_w);
+                    height = AV_CEIL_RSHIFT(height, desc->log2_chroma_h);
+                    pos_x  = AV_CEIL_RSHIFT(pos_x,  desc->log2_chroma_w);
+                    pos_y  = AV_CEIL_RSHIFT(pos_y,  desc->log2_chroma_h);
+                }
+
+                for (int y = 0; y < height; y++) {
+                    const uint8_t *src = frame->data[plane] + y * frame->linesize[plane];
+                    uint8_t *dst = canvas->data[plane] + (y + pos_y) * canvas->linesize[plane] + pos_x;
+                    memcpy(dst, src, width);
+                }
+            }
+
+            if (desc->nb_components < 4) {
+                // frame does not have alpha, set alpha to 255
+                desc = av_pix_fmt_desc_get(canvas->format);
+                plane  = desc->comp[3].plane;
+                width  = s->width;
+                height = s->height;
+                pos_x  = s->pos_x;
+                pos_y  = s->pos_y;
+
+                for (int y = 0; y < height; y++) {
+                    uint8_t *dst = canvas->data[plane] + (y + pos_y) * canvas->linesize[plane] + pos_x;
+                    memset(dst, 255, width);
+                }
+            }
+        }
+    } else {
+        // alpha blending
+
+        if (canvas->format == AV_PIX_FMT_ARGB) {
+            int src2_step[4] = { sizeof(uint32_t) };
+            blend_alpha_argb(canvas->data, canvas->linesize,
+                             (const uint8_t **) frame->data, frame->linesize,
+                             (const uint8_t **) canvas->data, canvas->linesize,
+                             src2_step, s->width, s->height, s->pos_x, s->pos_y);
+        } else /* if (canvas->format == AV_PIX_FMT_YUVA420P) */ {
+            int src2_step[4] = { 1, 1, 1, 1 };
+            blend_alpha_yuva(s, canvas->data, canvas->linesize,
+                             (const uint8_t **) frame->data, frame->linesize,
+                             frame->format,
+                             (const uint8_t **) canvas->data, canvas->linesize,
+                             src2_step, s->width, s->height, s->pos_x, s->pos_y);
+        }
     }
 
-    return avpkt->size;
+    return 0;
+}
+
+static int copy_canvas_to_frame(WebPContext *s, AVFrame *frame, int key_frame)
+{
+    AVFrame *canvas = s->canvas_frame.f;
+    int ret;
+
+    frame->format = canvas->format;
+    frame->width  = canvas->width;
+    frame->height = canvas->height;
+
+    ret = av_frame_get_buffer(frame, 0);
+    if (ret < 0)
+        return ret;
+
+    ret = av_frame_copy_props(frame, canvas);
+    if (ret < 0)
+        return ret;
+
+    // blend the canvas with the background color into the output frame
+    if (canvas->format == AV_PIX_FMT_ARGB) {
+        int src2_step[4] = { 0 };
+        const uint8_t *src2_data[4] = { &s->background_argb[0] };
+        blend_alpha_argb(frame->data, frame->linesize,
+                         (const uint8_t **) canvas->data, canvas->linesize,
+                         (const uint8_t **) src2_data, src2_step, src2_step,
+                         canvas->width, canvas->height, 0, 0);
+    } else /* if (canvas->format == AV_PIX_FMT_YUVA420P) */ {
+        int src2_step[4] = { 0, 0, 0, 0 };
+        blend_alpha_yuva(s, frame->data, frame->linesize,
+                         (const uint8_t **) canvas->data, canvas->linesize,
+                         canvas->format,
+                         s->background_data, src2_step, src2_step,
+                         canvas->width, canvas->height, 0, 0);
+    }
+
+    if (key_frame) {
+        frame->pict_type = AV_PICTURE_TYPE_I;
+    } else {
+        frame->pict_type = AV_PICTURE_TYPE_P;
+    }
+
+    return 0;
+}
+
+static int dispose_prev_frame_in_canvas(WebPContext *s)
+{
+    AVFrame *canvas = s->canvas_frame.f;
+    int width, height;
+    int pos_x, pos_y;
+
+    if ((s->prev_anmf_flags & ANMF_DISPOSAL_METHOD) == ANMF_DISPOSAL_METHOD_BACKGROUND) {
+        // dispose to background
+
+        if (canvas->format == AV_PIX_FMT_ARGB) {
+            width  = s->prev_width;
+            height = s->prev_height;
+            pos_x  = s->prev_pos_x;
+            pos_y  = s->prev_pos_y;
+
+            for (int y = 0; y < height; y++) {
+                uint32_t *dst = (uint32_t *) (canvas->data[0] + (y + pos_y) * canvas->linesize[0]) + pos_x;
+                memset(dst, 0, width * sizeof(uint32_t));
+            }
+        } else /* if (canvas->format == AV_PIX_FMT_YUVA420P) */ {
+            const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(canvas->format);
+            int plane;
+
+            for (int comp = 0; comp < desc->nb_components; comp++) {
+                plane  = desc->comp[comp].plane;
+                width  = s->prev_width;
+                height = s->prev_height;
+                pos_x  = s->prev_pos_x;
+                pos_y  = s->prev_pos_y;
+                if (comp == 1 || comp == 2) {
+                    width  = AV_CEIL_RSHIFT(width,  desc->log2_chroma_w);
+                    height = AV_CEIL_RSHIFT(height, desc->log2_chroma_h);
+                    pos_x  = AV_CEIL_RSHIFT(pos_x,  desc->log2_chroma_w);
+                    pos_y  = AV_CEIL_RSHIFT(pos_y,  desc->log2_chroma_h);
+                }
+
+                for (int y = 0; y < height; y++) {
+                    uint8_t *dst = canvas->data[plane] + (y + pos_y) * canvas->linesize[plane] + pos_x;
+                    memset(dst, s->transparent_yuva[plane], width);
+                }
+            }
+        }
+    }
+
+    return 0;
+}
+
+static int webp_decode_frame(AVCodecContext *avctx, AVFrame *p,
+                             int *got_frame, AVPacket *avpkt)
+{
+    WebPContext *s = avctx->priv_data;
+    AVFrame *canvas = s->canvas_frame.f;
+    int ret;
+    int key_frame = avpkt->flags & AV_PKT_FLAG_KEY;
+
+    *got_frame   = 0;
+
+    if (key_frame) {
+        // The canvas is passed from one thread to another in a sequence
+        // starting with a key frame followed by non-key frames.
+        // The key frame reports progress 1,
+        // the N-th non-key frame awaits progress N = s->await_progress
+        // and reports progress N + 1.
+        s->await_progress = 0;
+    }
+
+    // reset the frame params
+    s->anmf_flags = 0;
+    s->width      = 0;
+    s->height     = 0;
+    s->pos_x      = 0;
+    s->pos_y      = 0;
+    s->has_alpha  = 0;
+
+    ret = webp_decode_frame_common(avctx, avpkt->data, avpkt->size, got_frame, key_frame);
+    if (ret < 0)
+        goto end;
+
+    if (s->vp8x_flags & VP8X_FLAG_ANIMATION) {
+        // VP8 decoder might have changed the width and height of the frame
+        AVFrame *frame  = s->frame;
+        ret = av_frame_copy_props(canvas, frame);
+        if (ret < 0)
+            return ret;
+
+        ret = ff_set_dimensions(s->avctx, canvas->width, canvas->height);
+        if (ret < 0)
+            return ret;
+
+        s->avctx->pix_fmt = canvas->format;
+    }
+
+    ff_thread_finish_setup(s->avctx);
+
+    if (*got_frame) {
+        if (!(s->vp8x_flags & VP8X_FLAG_ANIMATION)) {
+            // no animation, output the decoded frame
+            av_frame_move_ref(p, s->frame);
+        } else {
+            if (!key_frame) {
+                ff_thread_await_progress(&s->canvas_frame, s->await_progress, 0);
+
+                ret = dispose_prev_frame_in_canvas(s);
+                if (ret < 0)
+                    goto end;
+            }
+
+            ret = blend_frame_into_canvas(s);
+            if (ret < 0)
+                goto end;
+
+            ret = copy_canvas_to_frame(s, p, key_frame);
+            if (ret < 0)
+                goto end;
+
+            ff_thread_report_progress(&s->canvas_frame, s->await_progress + 1, 0);
+        }
+
+        p->pts = avpkt->pts;
+    }
+
+    ret = avpkt->size;
+
+end:
+    av_frame_unref(s->frame);
+    return ret;
 }
 
 static av_cold int webp_decode_init(AVCodecContext *avctx)
 {
     WebPContext *s = avctx->priv_data;
+    const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(AV_PIX_FMT_YUVA420P);
 
+    s->avctx = avctx;
     s->pkt = av_packet_alloc();
-    if (!s->pkt)
+    s->canvas_frame.f = av_frame_alloc();
+    s->frame = av_frame_alloc();
+    if (!s->pkt || !s->canvas_frame.f || !s->frame) {
+        av_packet_free(&s->pkt);
+        av_frame_free(&s->canvas_frame.f);
+        av_frame_free(&s->frame);
         return AVERROR(ENOMEM);
+    }
+
+    // prepare data pointers for YUVA background
+    for (int i = 0; i < 4; i++)
+        s->background_data[i] = &s->background_yuva[i];
+
+    // convert transparent black from RGBA to YUVA
+    s->transparent_yuva[desc->comp[0].plane] = RGB_TO_Y_CCIR(0, 0, 0);
+    s->transparent_yuva[desc->comp[1].plane] = RGB_TO_U_CCIR(0, 0, 0, 0);
+    s->transparent_yuva[desc->comp[2].plane] = RGB_TO_V_CCIR(0, 0, 0, 0);
+    s->transparent_yuva[desc->comp[3].plane] = 0;
 
     return 0;
 }
@@ -1546,6 +2100,9 @@  static av_cold int webp_decode_close(AVCodecContext *avctx)
     WebPContext *s = avctx->priv_data;
 
     av_packet_free(&s->pkt);
+    ff_thread_release_ext_buffer(&s->canvas_frame);
+    av_frame_free(&s->canvas_frame.f);
+    av_frame_free(&s->frame);
 
     if (s->initialized)
         return ff_vp8_decode_free(avctx);
@@ -1553,15 +2110,56 @@  static av_cold int webp_decode_close(AVCodecContext *avctx)
     return 0;
 }
 
+static void webp_decode_flush(AVCodecContext *avctx)
+{
+    WebPContext *s = avctx->priv_data;
+
+    ff_thread_release_ext_buffer(&s->canvas_frame);
+}
+
+#if HAVE_THREADS
+static int webp_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
+{
+    WebPContext *wsrc = src->priv_data;
+    WebPContext *wdst = dst->priv_data;
+    int ret;
+
+    if (dst == src)
+        return 0;
+
+    ff_thread_release_ext_buffer(&wdst->canvas_frame);
+    if (wsrc->canvas_frame.f->data[0] &&
+        (ret = ff_thread_ref_frame(&wdst->canvas_frame, &wsrc->canvas_frame)) < 0)
+        return ret;
+
+    wdst->vp8x_flags      = wsrc->vp8x_flags;
+    wdst->canvas_width    = wsrc->canvas_width;
+    wdst->canvas_height   = wsrc->canvas_height;
+    wdst->prev_anmf_flags = wsrc->anmf_flags;
+    wdst->prev_width      = wsrc->width;
+    wdst->prev_height     = wsrc->height;
+    wdst->prev_pos_x      = wsrc->pos_x;
+    wdst->prev_pos_y      = wsrc->pos_y;
+    wdst->await_progress  = wsrc->await_progress + 1;
+
+    memcpy(wdst->background_argb,  wsrc->background_argb,  sizeof(wsrc->background_argb));
+    memcpy(wdst->background_yuva,  wsrc->background_yuva,  sizeof(wsrc->background_yuva));
+
+    return 0;
+}
+#endif
+
 const FFCodec ff_webp_decoder = {
     .p.name         = "webp",
     CODEC_LONG_NAME("WebP image"),
     .p.type         = AVMEDIA_TYPE_VIDEO,
     .p.id           = AV_CODEC_ID_WEBP,
     .priv_data_size = sizeof(WebPContext),
+    UPDATE_THREAD_CONTEXT(webp_update_thread_context),
     .init           = webp_decode_init,
     FF_CODEC_DECODE_CB(webp_decode_frame),
     .close          = webp_decode_close,
+    .flush          = webp_decode_flush,
     .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_FRAME_THREADS,
-    .caps_internal  = FF_CODEC_CAP_ICC_PROFILES,
+    .caps_internal  = FF_CODEC_CAP_ICC_PROFILES | FF_CODEC_CAP_ALLOCATE_PROGRESS,
 };