[FFmpeg-devel] [PATCH 1/3] doc/encoders.texi: Document cinepak encoder

[FFmpeg-devel] [PATCH 1/3] doc/encoders.texi: Document cinepak encoder

[Tomas Härdin]

[-- Attachment #1: Type: text/plain, Size: 1 bytes --]



[-- Attachment #2: 0001-doc-encoders.texi-Document-cinepak-encoder.patch --]
[-- Type: text/x-patch, Size: 2092 bytes --]

From 3cca353f177b9b91e1472fa60a6295e576142c59 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Tomas=20H=C3=A4rdin?= <git@haerdin.se>
Date: Sat, 9 Apr 2022 10:21:31 +0200
Subject: [PATCH 1/3] doc/encoders.texi: Document cinepak encoder

---
 doc/encoders.texi | 48 +++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 48 insertions(+)

diff --git a/doc/encoders.texi b/doc/encoders.texi
index 806cc430d4..1adfe382a4 100644
--- a/doc/encoders.texi
+++ b/doc/encoders.texi
@@ -1271,6 +1271,54 @@ follows.
 
 A64 / Commodore 64 multicolor charset encoder. @code{a64_multi5} is extended with 5th color (colram).
 
+@section Cinepak
+
+Cinepak aka CVID encoder.
+Compatible with Windows 3.1 and vintage MacOS.
+
+@subsection Options
+
+@table @option
+@item keyint_min @var{integer}
+Keyframe interval.
+A keyframe is inserted every @code{-keyint_min}-th frame.
+
+@item q:v @var{integer}
+Quality factor. Lower is better. Higher gives lower bitrate.
+The following table lists bitrates when encoding akiyo_cif.y4m for various values of @code{-q:v} with @code{-keyint_min 100}:
+
+@table @option
+@item @code{-q:v 1} 1918 kb/s
+@item @code{-q:v 2} 1735 kb/s
+@item @code{-q:v 4} 1500 kb/s
+@item @code{-q:v 10} 1064 kb/s
+@item @code{-q:v 20} 826 kb/s
+@item @code{-q:v 40} 553 kb/s
+@item @code{-q:v 100} 394 kb/s
+@item @code{-q:v 200} 311 kb/s
+@item @code{-q:v 400} 266 kb/s
+@item @code{-q:v 1000} 237 kb/s
+@end table
+
+@item max_extra_cb_iterations @var{integer}
+Max extra codebook recalculation passes, more is better and slower.
+
+@item skip_empty_cb @var{boolean}
+Avoid wasting bytes, ignore vintage MacOS decoder.
+
+@item max_strips @var{integer}
+@itemx min_strips @var{integer}
+The minimum and maximum number of strips to use.
+Wider range sometimes improves quality.
+More strips is generally better quality but costs more bits.
+Vintage compatible is 1..3.
+
+@item strip_number_adaptivity @var{integer}
+How much number of strips is allowed to change between frames.
+Higher is better but slower.
+
+@end table
+
 @section GIF
 
 GIF image/animation encoder.
-- 
2.30.2


[-- Attachment #3: Type: text/plain, Size: 251 bytes --]

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

[FFmpeg-devel] [PATCH 2/3] libavcodec/cinepakenc: Mark no-skip frames as keyframes

[Tomas Härdin]

[-- Attachment #1: Type: text/plain, Size: 1 bytes --]



[-- Attachment #2: 0002-libavcodec-cinepakenc-Mark-no-skip-frames-as-keyfram.patch --]
[-- Type: text/x-patch, Size: 10239 bytes --]

From 442bf0fbba4e13aa7d05ff5c30b07d4c94a8182c Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Tomas=20H=C3=A4rdin?= <git@haerdin.se>
Date: Sat, 9 Apr 2022 14:38:47 +0200
Subject: [PATCH 2/3] libavcodec/cinepakenc: Mark no-skip frames as keyframes

Reset curframe whenever we generate a keyframe.
Use -g instead of -keyint_min.
---
 libavcodec/cinepakenc.c              | 50 +++++++++++++++++++---------
 tests/ref/vsynth/vsynth1-cinepak     |  8 ++---
 tests/ref/vsynth/vsynth2-cinepak     |  8 ++---
 tests/ref/vsynth/vsynth_lena-cinepak |  6 ++--
 4 files changed, 46 insertions(+), 26 deletions(-)

diff --git a/libavcodec/cinepakenc.c b/libavcodec/cinepakenc.c
index 8b32c02780..6cfe8de200 100644
--- a/libavcodec/cinepakenc.c
+++ b/libavcodec/cinepakenc.c
@@ -113,7 +113,7 @@ typedef struct CinepakEncContext {
     enum AVPixelFormat pix_fmt;
     int w, h;
     int frame_buf_size;
-    int curframe, keyint;
+    int curframe;
     AVLFG randctx;
     uint64_t lambda;
     int *codebook_input;
@@ -215,7 +215,6 @@ static av_cold int cinepak_encode_init(AVCodecContext *avctx)
     s->h              = avctx->height;
     s->frame_buf_size = frame_buf_size;
     s->curframe       = 0;
-    s->keyint         = avctx->keyint_min;
     s->pix_fmt        = avctx->pix_fmt;
 
     // set up AVFrames
@@ -835,8 +834,7 @@ static void calculate_skip_errors(CinepakEncContext *s, int h,
         }
 }
 
-static void write_strip_header(CinepakEncContext *s, int y, int h, int keyframe,
-                               unsigned char *buf, int strip_size)
+static void write_strip_keyframe(unsigned char *buf, int keyframe)
 {
     // actually we are exclusively using intra strip coding (how much can we win
     // otherwise? how to choose which part of a codebook to update?),
@@ -844,6 +842,12 @@ static void write_strip_header(CinepakEncContext *s, int y, int h, int keyframe,
     // (besides, the logic here used to be inverted: )
     //    buf[0] = keyframe ? 0x11: 0x10;
     buf[0] = keyframe ? 0x10 : 0x11;
+}
+
+static void write_strip_header(CinepakEncContext *s, int y, int h, int keyframe,
+                               unsigned char *buf, int strip_size)
+{
+    write_strip_keyframe(buf, keyframe);
     AV_WB24(&buf[1], strip_size + STRIP_HEADER_SIZE);
     // AV_WB16(&buf[4], y); /* using absolute y values works -- rl */
     AV_WB16(&buf[4], 0); /* using relative values works as well -- rl */
@@ -857,7 +861,7 @@ static int rd_strip(CinepakEncContext *s, int y, int h, int keyframe,
                     uint8_t *last_data[4], int last_linesize[4],
                     uint8_t *data[4], int linesize[4],
                     uint8_t *scratch_data[4], int scratch_linesize[4],
-                    unsigned char *buf, int64_t *best_score)
+                    unsigned char *buf, int64_t *best_score, int *no_skip)
 {
     int64_t score = 0;
     int best_size = 0;
@@ -973,6 +977,9 @@ static int rd_strip(CinepakEncContext *s, int y, int h, int keyframe,
                                             scratch_data, scratch_linesize,
                                             last_data, last_linesize, &info,
                                             s->strip_buf + STRIP_HEADER_SIZE);
+                    // in theory we could have MODE_MC without ENC_SKIP,
+                    // but MODE_V1_V4 will always be more efficient
+                    *no_skip = info.mode != MODE_MC;
 
                     write_strip_header(s, y, h, keyframe, s->strip_buf, best_size);
                 }
@@ -999,13 +1006,13 @@ static int write_cvid_header(CinepakEncContext *s, unsigned char *buf,
 }
 
 static int rd_frame(CinepakEncContext *s, const AVFrame *frame,
-                    int isakeyframe, unsigned char *buf, int buf_size)
+                    int isakeyframe, unsigned char *buf, int buf_size, int *got_keyframe)
 {
     int num_strips, strip, i, y, nexty, size, temp_size, best_size;
     uint8_t *last_data    [4], *data    [4], *scratch_data    [4];
     int      last_linesize[4],  linesize[4],  scratch_linesize[4];
     int64_t best_score = 0, score, score_temp;
-    int best_nstrips;
+    int best_nstrips, best_strip_offsets[MAX_STRIPS];
 
     if (s->pix_fmt == AV_PIX_FMT_RGB24) {
         int x;
@@ -1064,12 +1071,15 @@ static int rd_frame(CinepakEncContext *s, const AVFrame *frame,
     // would be nice but quite certainly incompatible with vintage players:
     // support encoding zero strips (meaning skip the whole frame)
     for (num_strips = s->min_strips; num_strips <= s->max_strips && num_strips <= s->h / MB_SIZE; num_strips++) {
+        int strip_offsets[MAX_STRIPS];
+        int all_no_skip = 1;
         score = 0;
         size  = 0;
 
         for (y = 0, strip = 1; y < s->h; strip++, y = nexty) {
-            int strip_height;
+            int strip_height, no_skip;
 
+            strip_offsets[strip-1] = size + CVID_HEADER_SIZE;
             nexty = strip * s->h / num_strips; // <= s->h
             // make nexty the next multiple of 4 if not already there
             if (nexty & 3)
@@ -1099,21 +1109,24 @@ static int rd_frame(CinepakEncContext *s, const AVFrame *frame,
             if ((temp_size = rd_strip(s, y, strip_height, isakeyframe,
                                       last_data, last_linesize, data, linesize,
                                       scratch_data, scratch_linesize,
-                                      s->frame_buf + size + CVID_HEADER_SIZE,
-                                      &score_temp)) < 0)
+                                      s->frame_buf + strip_offsets[strip-1],
+                                      &score_temp, &no_skip)) < 0)
                 return temp_size;
 
             score += score_temp;
             size += temp_size;
+            all_no_skip &= no_skip;
         }
 
         if (best_score == 0 || score < best_score) {
             best_score = score;
-            best_size = size + write_cvid_header(s, s->frame_buf, num_strips, size, isakeyframe);
+            best_size = size + write_cvid_header(s, s->frame_buf, num_strips, size, all_no_skip);
 
             FFSWAP(AVFrame *, s->best_frame, s->scratch_frame);
             memcpy(buf, s->frame_buf, best_size);
             best_nstrips = num_strips;
+            *got_keyframe = all_no_skip; // no skip MBs in any strip -> keyframe
+            memcpy(best_strip_offsets, strip_offsets, sizeof(strip_offsets));
         }
         // avoid trying too many strip numbers without a real reason
         // (this makes the processing of the very first frame faster)
@@ -1121,6 +1134,11 @@ static int rd_frame(CinepakEncContext *s, const AVFrame *frame,
             break;
     }
 
+    // update strip headers
+    for (i = 0; i < best_nstrips; i++) {
+        write_strip_keyframe(s->frame_buf + best_strip_offsets[i], *got_keyframe);
+    }
+
     // let the number of strips slowly adapt to the changes in the contents,
     // compared to full bruteforcing every time this will occasionally lead
     // to some r/d performance loss but makes encoding up to several times faster
@@ -1151,21 +1169,23 @@ static int cinepak_encode_frame(AVCodecContext *avctx, AVPacket *pkt,
                                 const AVFrame *frame, int *got_packet)
 {
     CinepakEncContext *s = avctx->priv_data;
-    int ret;
+    int ret, got_keyframe;
 
     s->lambda = frame->quality ? frame->quality - 1 : 2 * FF_LAMBDA_SCALE;
 
     if ((ret = ff_alloc_packet(avctx, pkt, s->frame_buf_size)) < 0)
         return ret;
-    ret       = rd_frame(s, frame, (s->curframe == 0), pkt->data, s->frame_buf_size);
+    ret       = rd_frame(s, frame, (s->curframe == 0), pkt->data, s->frame_buf_size, &got_keyframe);
     pkt->size = ret;
-    if (s->curframe == 0)
+    if (got_keyframe) {
         pkt->flags |= AV_PKT_FLAG_KEY;
+        s->curframe = 0;
+    }
     *got_packet = 1;
 
     FFSWAP(AVFrame *, s->last_frame, s->best_frame);
 
-    if (++s->curframe >= s->keyint)
+    if (++s->curframe >= avctx->gop_size)
         s->curframe = 0;
 
     return 0;
diff --git a/tests/ref/vsynth/vsynth1-cinepak b/tests/ref/vsynth/vsynth1-cinepak
index e47ae26b6d..7cb33220e7 100644
--- a/tests/ref/vsynth/vsynth1-cinepak
+++ b/tests/ref/vsynth/vsynth1-cinepak
@@ -1,4 +1,4 @@
-cd28e47a6ac396240a3fee69f15625d1 *tests/data/fate/vsynth1-cinepak.avi
-408616 tests/data/fate/vsynth1-cinepak.avi
-e74066a028c708f467272884ecd3f7d3 *tests/data/fate/vsynth1-cinepak.out.rawvideo
-stddev:   61.38 PSNR: 12.37 MAXDIFF:  225 bytes:  7603200/   921600
+6b9b4d2b32854fd8b0f84ee4d76b1fed *tests/data/fate/vsynth1-cinepak.avi
+408910 tests/data/fate/vsynth1-cinepak.avi
+4acdcc3e664d12facbcefc389079cfa6 *tests/data/fate/vsynth1-cinepak.out.rawvideo
+stddev:   61.38 PSNR: 12.37 MAXDIFF:  223 bytes:  7603200/   921600
diff --git a/tests/ref/vsynth/vsynth2-cinepak b/tests/ref/vsynth/vsynth2-cinepak
index 835de78d16..72b8794116 100644
--- a/tests/ref/vsynth/vsynth2-cinepak
+++ b/tests/ref/vsynth/vsynth2-cinepak
@@ -1,4 +1,4 @@
-663a2804f421709208c76f6e34e7bea5 *tests/data/fate/vsynth2-cinepak.avi
-400402 tests/data/fate/vsynth2-cinepak.avi
-2c761c3c8cda083eb8f54b2df72b257b *tests/data/fate/vsynth2-cinepak.out.rawvideo
-stddev:   80.96 PSNR:  9.96 MAXDIFF:  227 bytes:  7603200/   921600
+675d6a4370c315dc53c219a9271be4c6 *tests/data/fate/vsynth2-cinepak.avi
+400720 tests/data/fate/vsynth2-cinepak.avi
+cbf6900b882b484bc531f9b4799c226e *tests/data/fate/vsynth2-cinepak.out.rawvideo
+stddev:   80.96 PSNR:  9.96 MAXDIFF:  229 bytes:  7603200/   921600
diff --git a/tests/ref/vsynth/vsynth_lena-cinepak b/tests/ref/vsynth/vsynth_lena-cinepak
index 6b021673a0..c9cb2cfaba 100644
--- a/tests/ref/vsynth/vsynth_lena-cinepak
+++ b/tests/ref/vsynth/vsynth_lena-cinepak
@@ -1,4 +1,4 @@
-a9ea19eb0d239a53af8630d5bc4167d0 *tests/data/fate/vsynth_lena-cinepak.avi
-407574 tests/data/fate/vsynth_lena-cinepak.avi
-e32d4103194665d2ea0f46d5cdd0cdf2 *tests/data/fate/vsynth_lena-cinepak.out.rawvideo
+a79f907fecb95c8a5b5de670ccda215b *tests/data/fate/vsynth_lena-cinepak.avi
+407952 tests/data/fate/vsynth_lena-cinepak.avi
+2664cdff5f14508e891477b498cc840e *tests/data/fate/vsynth_lena-cinepak.out.rawvideo
 stddev:   58.10 PSNR: 12.85 MAXDIFF:  185 bytes:  7603200/   921600
-- 
2.30.2


[-- Attachment #3: Type: text/plain, Size: 251 bytes --]

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

[FFmpeg-devel] [PATCH 3/3] doc/encoders.texi: Document updated behavior

[Tomas Härdin]

[-- Attachment #1: Type: text/plain, Size: 262 bytes --]

I forgot to comment on the previous patch the change from -keyint_min
to -g. It didn't really make much difference then since the encoder
would put keyframes at fixed interval in addition to any accidental
keyframes that the encoder may deem appropriate

/Tomas

[-- Attachment #2: 0003-doc-encoders.texi-Document-updated-behavior.patch --]
[-- Type: text/x-patch, Size: 1912 bytes --]

From 99c3b511252ba403b1aa7f00d31f9455304c50da Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Tomas=20H=C3=A4rdin?= <git@haerdin.se>
Date: Sat, 9 Apr 2022 14:40:47 +0200
Subject: [PATCH 3/3] doc/encoders.texi: Document updated behavior

Bitrates of akiyo changed slightly.
---
 doc/encoders.texi | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/doc/encoders.texi b/doc/encoders.texi
index 1adfe382a4..75e49ec359 100644
--- a/doc/encoders.texi
+++ b/doc/encoders.texi
@@ -1279,23 +1279,23 @@ Compatible with Windows 3.1 and vintage MacOS.
 @subsection Options
 
 @table @option
-@item keyint_min @var{integer}
+@item g @var{integer}
 Keyframe interval.
-A keyframe is inserted every @code{-keyint_min}-th frame.
+A keyframe is inserted at least every @code{-g} frames, sometimes sooner.
 
 @item q:v @var{integer}
 Quality factor. Lower is better. Higher gives lower bitrate.
-The following table lists bitrates when encoding akiyo_cif.y4m for various values of @code{-q:v} with @code{-keyint_min 100}:
+The following table lists bitrates when encoding akiyo_cif.y4m for various values of @code{-q:v} with @code{-g 100}:
 
 @table @option
 @item @code{-q:v 1} 1918 kb/s
 @item @code{-q:v 2} 1735 kb/s
 @item @code{-q:v 4} 1500 kb/s
-@item @code{-q:v 10} 1064 kb/s
+@item @code{-q:v 10} 1041 kb/s
 @item @code{-q:v 20} 826 kb/s
 @item @code{-q:v 40} 553 kb/s
 @item @code{-q:v 100} 394 kb/s
-@item @code{-q:v 200} 311 kb/s
+@item @code{-q:v 200} 312 kb/s
 @item @code{-q:v 400} 266 kb/s
 @item @code{-q:v 1000} 237 kb/s
 @end table
@@ -1311,6 +1311,7 @@ Avoid wasting bytes, ignore vintage MacOS decoder.
 The minimum and maximum number of strips to use.
 Wider range sometimes improves quality.
 More strips is generally better quality but costs more bits.
+Fewer strips tend to yield more keyframes.
 Vintage compatible is 1..3.
 
 @item strip_number_adaptivity @var{integer}
-- 
2.30.2


[-- Attachment #3: Type: text/plain, Size: 251 bytes --]

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

Re: [FFmpeg-devel] [PATCH 1/3] doc/encoders.texi: Document cinepak encoder

[Tomas Härdin]

Ping

I tested patch 2's output on Windows 3.1, still works fine. Could
potentially merge both documentation patches

/Tomas



_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

Re: [FFmpeg-devel] [PATCH 1/3] doc/encoders.texi: Document cinepak encoder

[Tomas Härdin]

ons 2022-05-04 klockan 11:41 +0200 skrev Tomas Härdin:
> Ping
> 
> I tested patch 2's output on Windows 3.1, still works fine. Could
> potentially merge both documentation patches

Pushing this in a day or two. FATE passes.

/Tomas


_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

Re: [FFmpeg-devel] [PATCH 1/3] doc/encoders.texi: Document cinepak encoder

[Tomas Härdin]

tis 2022-05-10 klockan 09:59 +0200 skrev Tomas Härdin:
> ons 2022-05-04 klockan 11:41 +0200 skrev Tomas Härdin:
> > Ping
> > 
> > I tested patch 2's output on Windows 3.1, still works fine. Could
> > potentially merge both documentation patches
> 
> Pushing this in a day or two. FATE passes.

Pushed

/Tomas

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".