feat: seek() filter and ffmpeg.SeekPosition, ffmpeg.SeekDuration

* feat: seek() filter and ffmpeg.SeekPosition, ffmpeg.SeekDuration

README.md

@@ -41,12 +41,34 @@ imagorvideo then converts the selected frame to RGB image data, forwards to the
 
 imagorvideo supports the following filters, which can be used in conjunction with [imagor filters](https://github.com/cshum/imagor#filters):
 
-- `frame(n)` specifying the time position, duration or frame index for imaging, which skips the default automatic selection:
+- `frame(n)` specify the position or time duration for imaging, which skips the automatic best frame selection:
   - Float between `0.0` and `1.0` indices position of the video. Example `frame(0.5)`, `frame(1.0)`
   - Time duration indices the elasped time since the start of video. Example `frame(5m1s)`, `frame(200s)`
-  - Number starts from 1 indices frame index, example `frame(1)`, `frame(10)`
+- `seek(n)` seeks to the approximate position or time duration, then perform automatic best frame selection around that point:
+  - Float between `0.0` and `1.0` indices position of the video. Example `seek(0.5)`
+  - Time duration indices the elasped time since the start of video. Example `seek(5m1s)`, `seek(200s)`
 - `max_frames(n)` restrict the maximum number of frames allocated for image selection. The smaller the number, the faster the processing time.
 
+#### `frame(n)` vs `seek(n)`
+
+There are differences you may want to choose one over the other.
+`frame(n)` gives you the precise time frame specified. However, precise may not be the best in some circumstances:
+```
+http://commondatastorage.googleapis.com/gtv-videos-bucket/sample/BigBuckBunny.mp4
+```
+Retrieving the frame at 5 minutes elapsed time of this video:
+```
+http://localhost:8000/unsafe/300x0/filters:frame(5m)/http://commondatastorage.googleapis.com/gtv-videos-bucket/sample/BigBuckBunny.mp4
+```
+It results a complete black frame. 
+
+![black](https://raw.githubusercontent.com/cshum/imagorvideo/master/testdata/black.jpg)
+
+This is where `seek(n)` comes handy. It seeks to the key frame before the 5 minutes elapsed time, then perform best frame selection starting from that point using root-mean-square error (RMSE).
+The result is a reasonable image that sits close to the specified time:
+
+![seek 5m](https://raw.githubusercontent.com/cshum/imagorvideo/master/testdata/seek5m.jpg)
+
 ### Metadata
 
 imagorvideo provides metadata endpoint that extracts video metadata, including dimension, duration and FPS data. It processes header only, without extracting the frame data for better processing speed.

ffmpeg/ffmpeg.go

@@ -107,13 +107,21 @@ func (av *AVContext) SelectPosition(f float64) (err error) {
 func (av *AVContext) SelectDuration(ts time.Duration) (err error) {
 	if ts > 0 {
 		av.selectedDuration = ts
-		if err = seekDuration(av, ts); err != nil {
+		if err = av.SeekDuration(ts); err != nil {
 			return
 		}
 	}
 	return av.ProcessFrames(-1)
 }
 
+func (av *AVContext) SeekPosition(f float64) error {
+	return av.SeekDuration(av.positionToDuration(f))
+}
+
+func (av *AVContext) SeekDuration(ts time.Duration) error {
+	return seekDuration(av, ts)
+}
+
 func (av *AVContext) Export(bands int) (buf []byte, err error) {
 	if err = av.ProcessFrames(-1); err != nil {
 		return

ffmpeg/ffmpeg_test.go

@@ -67,17 +67,6 @@ func TestAVContext(t *testing.T) {
 				stats, err := os.Stat(path)
 				require.NoError(t, err)
 				av, err := LoadAVContext(reader, stats.Size())
-				require.NoError(t, err)
-				defer av.Close()
-				if n == 10 {
-					require.NoError(t, av.ProcessFrames(n))
-				} else if n == 99999 {
-					require.NoError(t, av.SelectDuration(time.Second))
-				} else if n == 9999 {
-					require.NoError(t, av.SelectPosition(0.7))
-				} else if n > -1 {
-					require.NoError(t, av.SelectFrame(n))
-				}
 				meta := av.Metadata()
 				metaBuf, err := json.Marshal(meta)
 				require.NoError(t, err)
@@ -87,6 +76,19 @@ func TestAVContext(t *testing.T) {
 				} else {
 					require.NoError(t, os.WriteFile(goldenFile, metaBuf, 0666))
 				}
+				require.NoError(t, err)
+				defer av.Close()
+				if n == 10 {
+					require.NoError(t, av.ProcessFrames(n))
+				} else if n == 99999 {
+					require.NoError(t, av.SelectDuration(time.Second))
+				} else if n == 9999 {
+					require.NoError(t, av.SelectPosition(0.7))
+				} else if n == 5 {
+					require.NoError(t, av.SelectFrame(n))
+				} else {
+					require.NoError(t, av.SeekPosition(0.7))
+				}
 				bands := 4
 				if n == 99999 {
 					bands = 999

processor.go

@@ -136,6 +136,16 @@ func (p *Processor) Process(ctx context.Context, in *imagor.Blob, params imagorp
 					}
 				}
 			}
+		case "seek":
+			if ts, e := time.ParseDuration(filter.Args); e == nil {
+				if err = av.SeekDuration(ts); err != nil {
+					return
+				}
+			} else if f, e := strconv.ParseFloat(filter.Args, 64); e == nil {
+				if err = av.SeekPosition(f); err != nil {
+					return
+				}
+			}
 		case "max_frames":
 			n, _ := strconv.Atoi(filter.Args)
 			if err = av.ProcessFrames(n); err != nil {

processor_test.go

@@ -32,7 +32,6 @@ type test struct {
 	name       string
 	path       string
 	expectCode int
-	sizeOnly   bool
 }
 
 func TestProcessor(t *testing.T) {
@@ -55,8 +54,10 @@ func TestProcessor(t *testing.T) {
 		{name: "mp4 orient 270", path: "200x100/schizo_270.mp4"},
 		{name: "image", path: "fit-in/100x100/demo.png"},
 		{name: "alpha", path: "fit-in/filters:format(png)/alpha-webm.webm"},
-		{name: "alpha frame duration", path: "500x/filters:frame(5s):format(png)/alpha-webm.webm", sizeOnly: true},
-		{name: "alpha frame position", path: "500x/filters:frame(0.5):format(png)/alpha-webm.webm", sizeOnly: true},
+		{name: "alpha frame duration", path: "500x/filters:frame(5s):format(png)/alpha-webm.webm"},
+		{name: "alpha frame position", path: "500x/filters:frame(0.5):format(png)/alpha-webm.webm"},
+		{name: "alpha seek duration", path: "500x/filters:seek(5s):format(png)/alpha-webm.webm"},
+		{name: "alpha seek position", path: "500x/filters:seek(0.5):format(png)/alpha-webm.webm"},
 		{name: "corrupted", path: "fit-in/100x100/corrupt/everybody-betray-me.mkv", expectCode: 406},
 		{name: "no cover meta", path: "meta/no_cover.mp3"},
 		{name: "no cover 406", path: "fit-in/100x100/no_cover.mp3", expectCode: 406},
@@ -108,9 +109,6 @@ func doGoldenTests(t *testing.T, resultDir string, tests []test, opts ...Option)
 			assert.NoError(t, app.Shutdown(context.Background()))
 		})
 		for _, tt := range tests {
-			if i == 1 && tt.sizeOnly {
-				continue
-			}
 			t.Run(fmt.Sprintf("%s-%d", tt.name, i+1), func(t *testing.T) {
 				w := httptest.NewRecorder()
 				ctx, cancel := context.WithCancel(context.Background())