fix(encoder): fix mixture

gnes/encoder/__init__.py

@@ -34,13 +34,15 @@
     'BaseEncoder': 'base',
     'BaseBinaryEncoder': 'base',
     'BaseTextEncoder': 'base',
+    'BaseVideoEncoder': 'base',
     'BaseNumericEncoder': 'base',
     'CompositionalEncoder': 'base',
     'PipelineEncoder': 'base',
     'HashEncoder': 'numeric.hash',
     'BasePytorchEncoder': 'image.base',
     'TFInceptionEncoder': 'image.inception',
-    'CVAEEncoder': 'image.cvae'
+    'CVAEEncoder': 'image.cvae',
+    'IncepMixtureEncoder': 'video.incep_mixture'
 }
 
 register_all_class(_cls2file_map, 'encoder')

gnes/encoder/base.py

@@ -38,6 +38,12 @@ def encode(self, img: List['np.ndarray'], *args, **kwargs) -> np.ndarray:
         pass
 
 
+class BaseVideoEncoder(BaseEncoder):
+
+    def encode(self, img: List['np.ndarray'], *args, **kwargs) -> np.ndarray:
+        pass
+
+
 class BaseTextEncoder(BaseEncoder):
 
     def encode(self, text: List[str], *args, **kwargs) -> np.ndarray:

gnes/preprocessor/__init__.py

@@ -30,6 +30,7 @@
     'BaseUnaryPreprocessor': 'base',
     'BaseVideoPreprocessor': 'video.base',
     'FFmpegPreprocessor': 'video.ffmpeg',
+    'FFmpegVideoSegmentor': 'video.ffmpeg',
     'ShotDetectPreprocessor': 'video.shotdetect',
 }
 

gnes/preprocessor/video/ffmpeg.py

@@ -106,3 +106,45 @@ def duplicate_rm_hash(self,
                 ret.append((i, h))
 
         return [images[_[0]] for _ in ret]
+
+
+class FFmpegVideoSegmentor(BaseVideoPreprocessor):
+    def __init__(self,
+                 frame_size: str = "192*168",
+                 segment_method: str = 'uniform',
+                 segment_interval: int = -1,
+                 *args,
+                 **kwargs):
+        super().__init__(*args, **kwargs)
+        self.frame_size = frame_size
+        self.segment_method = segment_method
+        self.segment_interval = segment_interval
+        self._ffmpeg_kwargs = kwargs
+
+    def apply(self, doc: 'gnes_pb2.Document') -> None:
+        super().apply(doc)
+        if doc.raw_bytes:
+            frames = get_video_frames(
+                doc.raw_bytes,
+                s=self.frame_size,
+                vsync=self._ffmpeg_kwargs.get("vsync", "vfr"),
+                vf=self._ffmpeg_kwargs.get("vf", "select=eq(pict_type\\,I)"))
+
+            sub_videos = []
+            if len(frames) >= 1:
+                if self.segment_method == 'uniform':
+                    if self.segment_interval == -1:
+                        sub_videos = [frames]
+                    else:
+                        sub_videos = [frames[_: _+self.segment_interval]
+                                      for _ in range(0, len(frames), self.segment_interval)]
+                    for ci, chunk in enumerate(sub_videos):
+                        c = doc.chunks.add()
+                        c.doc_id = doc.doc_id
+                        c.blob.CopyFrom(array2blob(np.array(chunk, dtype=np.uint8)))
+                        c.offset_1d = ci
+                        c.weight = 1 / len(sub_videos)
+            else:
+                self.logger.info('bad document: no key frames extracted')
+        else:
+            self.logger.error('bad document: "raw_bytes" is empty!')