yta-video-opengl 0.0.9__py3-none-any.whl → 0.0.11__py3-none-any.whl

yta_video_opengl/nodes/video/opengl.py

@@ -0,0 +1,309 @@
+from yta_video_opengl.utils import frame_to_texture, get_fullscreen_quad_vao
+from yta_video_opengl.nodes import Node
+from yta_validation.parameter import ParameterValidator
+from yta_validation import PythonValidator
+from abc import abstractmethod
+from typing import Union
+
+import numpy as np
+import moderngl
+
+
+class _Uniforms:
+    """
+    Class to wrap the functionality related to
+    handling the opengl program uniforms.
+    """
+
+    @property
+    def uniforms(
+        self
+    ) -> dict:
+        """
+        The uniforms in the program, as a dict, in
+        the format `{key, value}`.
+        """
+        return {
+            key: self.program[key].value
+            for key in self.program
+            if PythonValidator.is_instance_of(self.program[key], moderngl.Uniform)
+        }
+
+    def __init__(
+        self,
+        program: moderngl.Program
+    ):
+        self.program: moderngl.Program = program
+        """
+        The program instance this handler class
+        belongs to.
+        """
+
+    def get(
+        self,
+        name: str
+    ) -> Union[any, None]:
+        """
+        Get the value of the uniform with the
+        given 'name'.
+        """
+        return self.uniforms.get(name, None)
+
+    # TODO: I need to refactor these method to
+    # accept a **kwargs maybe, or to auto-detect
+    # the type and add the uniform as it must be
+    # done
+    def set(
+        self,
+        name: str,
+        value
+    ) -> '_Uniforms':
+        """
+        Set the provided 'value' to the normal type
+        uniform with the given 'name'. Here you have
+        some examples of defined uniforms we can set
+        with this method:
+        - `uniform float name;`
+
+        TODO: Add more examples
+        """
+        if name in self.program:
+            self.program[name].value = value
+
+        return self
+    
+    def set_vec(
+        self,
+        name: str,
+        values
+    ) -> '_Uniforms':
+        """
+        Set the provided 'value' to the normal type
+        uniform with the given 'name'. Here you have
+        some examples of defined uniforms we can set
+        with this method:
+        - `uniform vec2 name;`
+
+        TODO: Is this example ok? I didn't use it yet
+        """
+        if name in self.program:
+            self.program[name].write(np.array(values, dtype = 'f4').tobytes())
+
+        return self
+
+    def set_mat(
+        self,
+        name: str,
+        value
+    ) -> '_Uniforms':
+        """
+        Set the provided 'value' to a `matN` type
+        uniform with the given 'name'. The 'value'
+        must be a NxN matrix (maybe numpy array)
+        transformed to bytes ('.tobytes()').
+        
+        This uniform must be defined in the vertex
+        like this:
+        - `uniform matN name;`
+
+        TODO: Maybe we can accept a NxN numpy 
+        array and do the .tobytes() by ourselves...
+        """
+        if name in self.program:
+            self.program[name].write(value)
+
+        return self
+    
+    def print(
+        self
+    ) -> '_Uniforms':
+        """
+        Print the defined uniforms in console.
+        """
+        for key, value in self.uniforms.items():
+            print(f'"{key}": {str(value)}')
+
+class OpenglNode(Node):
+    """
+    The basic class of a node to manipulate frames
+    as opengl textures. This node will process the
+    frame as an input texture and will generate 
+    also a texture as the output.
+
+    Nodes can be chained and the result from one
+    node can be applied on another node.
+    """
+
+    @property
+    @abstractmethod
+    def vertex_shader(
+        self
+    ) -> str:
+        """
+        The code of the vertex shader.
+        """
+        pass
+
+    @property
+    @abstractmethod
+    def fragment_shader(
+        self
+    ) -> str:
+        """
+        The code of the fragment shader.
+        """
+        pass
+
+    def __init__(
+        self,
+        context: moderngl.Context,
+        size: tuple[int, int],
+        **kwargs
+    ):
+        ParameterValidator.validate_mandatory_instance_of('context', context, moderngl.Context)
+        # TODO: Validate size
+
+        self.context: moderngl.Context = context
+        """
+        The context of the program.
+        """
+        self.size: tuple[int, int] = size
+        """
+        The size we want to use for the frame buffer
+        in a (width, height) format.
+        """
+        # Compile shaders within the program
+        self.program: moderngl.Program = self.context.program(
+            vertex_shader = self.vertex_shader,
+            fragment_shader = self.fragment_shader
+        )
+
+        # Create the fullscreen quad
+        self.quad = get_fullscreen_quad_vao(
+            context = self.context,
+            program = self.program
+        )
+
+        # Create the output fbo
+        self.output_tex = self.context.texture(self.size, 4)
+        self.output_tex.filter = (moderngl.LINEAR, moderngl.LINEAR)
+        self.fbo = self.context.framebuffer(color_attachments = [self.output_tex])
+
+        self.uniforms: _Uniforms = _Uniforms(self.program)
+        """
+        Shortcut to the uniforms functionality.
+        """
+        # Auto set uniforms dynamically if existing
+        for key, value in kwargs.items():
+            self.uniforms.set(key, value)
+
+    def process(
+        self,
+        input: Union[moderngl.Texture, 'VideoFrame', 'np.ndarray']
+    ) -> moderngl.Texture:
+        """
+        Apply the shader to the 'input', that
+        must be a frame or a texture, and return
+        the new resulting texture.
+
+        We use and return textures to maintain
+        the process in GPU and optimize it.
+        """
+        if PythonValidator.is_instance_of(input, ['VideoFrame', 'ndarray']):
+            # TODO: What about the numpy format (?)
+            input = frame_to_texture(input, self.context)
+
+        self.fbo.use()
+        self.context.clear(0.0, 0.0, 0.0, 0.0)
+
+        input.use(location = 0)
+
+        if 'texture' in self.program:
+            self.program['texture'] = 0
+
+        self.quad.render()
+
+        return self.output_tex
+    
+class WavingNode(OpenglNode):
+    """
+    Just an example, without the shaders code
+    actually, to indicate that we can use
+    custom parameters to make it work.
+    """
+
+    @property
+    def vertex_shader(
+        self
+    ) -> str:
+        return (
+            '''
+            #version 330
+            in vec2 in_vert;
+            in vec2 in_texcoord;
+            out vec2 v_uv;
+            void main() {
+                v_uv = in_texcoord;
+                gl_Position = vec4(in_vert, 0.0, 1.0);
+            }
+            '''
+        )
+
+    @property
+    def fragment_shader(
+        self
+    ) -> str:
+        return (
+            '''
+            #version 330
+            uniform sampler2D tex;
+            uniform float time;
+            uniform float amplitude;
+            uniform float frequency;
+            uniform float speed;
+            in vec2 v_uv;
+            out vec4 f_color;
+            void main() {
+                float wave = sin(v_uv.x * frequency + time * speed) * amplitude;
+                vec2 uv = vec2(v_uv.x, v_uv.y + wave);
+                f_color = texture(tex, uv);
+            }
+            '''
+        )
+
+    def __init__(
+        self,
+        context: moderngl.Context,
+        size: tuple[int, int],
+        amplitude: float = 0.05,
+        frequency: float = 10.0,
+        speed: float = 2.0
+    ):
+        super().__init__(
+            context = context,
+            size = size,
+            amplitude = amplitude,
+            frequency = frequency,
+            speed = speed
+        )
+
+    # This is just an example and we are not
+    # using the parameters actually, but we
+    # could set those specific uniforms to be
+    # processed by the code
+    def process(
+        self,
+        input: Union[moderngl.Texture, 'VideoFrame', 'np.ndarray'],
+        t: float = 0.0,
+    ) -> moderngl.Texture:
+        """
+        Apply the shader to the 'input', that
+        must be a frame or a texture, and return
+        the new resulting texture.
+
+        We use and return textures to maintain
+        the process in GPU and optimize it.
+        """
+        self.uniforms.set('time', t)
+
+        return super().process(input)

yta_video_opengl/reader/__init__.py

@@ -575,6 +575,16 @@ class VideoReader:
         """
         return self.video_cache.get_frame(index)
     
+    def get_frame_from_t(
+        self,
+        t: float
+    ) -> 'VideoFrame':
+        """
+        Get the video frame with the given 't' time
+        moment, using the video cache system.
+        """
+        return self.video_cache.get_frame_from_t(t)
+    
     # TODO: Will we use this (?)
     def get_audio_frame(
         self,

yta_video_opengl/reader/cache.py

@@ -16,17 +16,22 @@ memory all those frames to be handled fast. It
 will remove the old frames if needed to use only
 the 'size' we set when creating it.
 """
-from yta_video_opengl.utils import t_to_pts, pts_to_t, pts_to_index
+from yta_video_opengl.utils import t_to_pts, pts_to_t, pts_to_index, index_to_pts
+from yta_video_frame_time import T
 from av.container import InputContainer
 from av.video.stream import VideoStream
 from av.audio.stream import AudioStream
 from av.video.frame import VideoFrame
 from av.audio.frame import AudioFrame
 from yta_validation.parameter import ParameterValidator
+from yta_validation import PythonValidator
 from fractions import Fraction
 from collections import OrderedDict
 from typing import Union
 
+import numpy as np
+import math
+
 
 class VideoFrameCache:
     """
@@ -60,11 +65,11 @@ class VideoFrameCache:
         self,
         container: InputContainer,
         stream: Union[VideoStream, AudioStream],
-        size: int = 50
+        size: Union[int, None] = None
     ):
         ParameterValidator.validate_mandatory_instance_of('container', container, InputContainer)
         ParameterValidator.validate_mandatory_instance_of('stream', stream, [VideoStream, AudioStream])
-        ParameterValidator.validate_mandatory_positive_int('size', size)
+        ParameterValidator.validate_positive_int('size', size)
 
         self.container: InputContainer = container
         """
@@ -78,7 +83,7 @@ class VideoFrameCache:
         """
         The cache ordered dictionary.
         """
-        self.size = size
+        self.size: Union[int, None] = size
         """
         The size (in number of frames) of the cache.
         """
@@ -99,6 +104,31 @@ class VideoFrameCache:
             if packet.is_keyframe:
                 self.key_frames_pts.append(packet.pts)
 
+        # The cache size will be auto-calculated to
+        # use the amount of frames of the biggest
+        # interval of frames that belongs to a key
+        # frame, or a value by default
+        fps = (
+            float(self.stream.average_rate)
+            if PythonValidator.is_instance_of(self.stream, VideoStream) else
+            float(self.stream.rate)
+        )
+        # Intervals, but in number of frames
+        intervals = np.diff(
+            # Intervals of time between keyframes
+            np.array(self.key_frames_pts) * self.stream.time_base
+        ) * fps
+
+        self.size = (
+            math.ceil(np.max(intervals))
+            if intervals.size > 0 else
+            (
+                self.size or
+                # TODO: Make this 'default_size' a setting or something
+                60
+            )
+        )
+        
         self.container.seek(0)
 
     def _get_nearest_keyframe_fps(
@@ -117,10 +147,29 @@ class VideoFrameCache:
             if key_frame_pts <= pts
         ])
 
+    def _store_frame_in_cache(
+        self,
+        frame: Union[VideoFrame, AudioFrame]
+    ) -> Union[VideoFrame, AudioFrame]:
+        """
+        Store the provided 'frame' in cache if it
+        is not on it, removing the first item of
+        the cache if full.
+        """
+        if frame.pts not in self.cache:
+            # TODO: The 'format' must be dynamic
+            self.cache[frame.pts] = frame
+
+            # Clean cache if full
+            if len(self.cache) > self.size:
+                self.cache.popitem(last = False)
+
+        return frame
+
     def _get_frame_by_pts(
         self,
         pts: int
-    ):
+    ) -> Union[VideoFrame, AudioFrame, None]:
         """
         Get the frame that has the provided 'pts'.
 
@@ -145,18 +194,14 @@ class VideoFrameCache:
                 continue
 
             # Store in cache if needed
-            if frame.pts not in self.cache:
-                # TODO: The 'format' must be dynamic
-                self.cache[frame.pts] = frame.to_ndarray(format = "rgb24")
-
-                # Clean cache if full
-                if len(self.cache) > self.size:
-                    self.cache.popitem(last = False)
+            self._store_frame_in_cache(frame)
 
             if frame.pts >= pts:
                 decoded = self.cache[frame.pts]
                 break
 
+        # TODO: Is this working? We need previous 
+        # frames to be able to decode...
         return decoded
 
     def get_frame(
@@ -167,14 +212,24 @@ class VideoFrameCache:
         Get the frame with the given 'index' from
         the cache.
         """
-        # TODO: Maybe we can accept 't' and 'pts' also
-        target_pts = int(index / self.fps / self.time_base)
+        # TODO: Maybe we can accept 'pts' also
+        pts = index_to_pts(index, self.time_base, self.fps)
 
         return (
-            self.cache[target_pts]
-            if target_pts in self.cache else
-            self._get_frame_by_pts(target_pts)
+            self.cache[pts]
+            if pts in self.cache else
+            self._get_frame_by_pts(pts)
         )
+    
+    def get_frame_from_t(
+        self,
+        t: float
+    ) -> Union[VideoFrame, AudioFrame]:
+        """
+        Get the frame with the given 't' time moment
+        from the cache.
+        """
+        return self.get_frame(T.video_frame_time_to_video_frame_index(t, self.fps))
 
     def get_frames(
         self,
@@ -186,10 +241,22 @@ class VideoFrameCache:
         the provided 'start' and 'end' time in
         seconds.
         """
-        # TODO: I create this method by default using
-        # the cache. Think about how to implement it
-        # and apply it here, please.
-        # Go to the nearest key frame
+        # We use the cache as iterator if all the frames
+        # requested are stored there
+        pts_list = [
+            t_to_pts(t, self.time_base)
+            for t in T.get_frame_indexes(self.stream.duration, self.fps, start, end)
+        ]
+
+        if all(
+            pts in self.cache
+            for pts in pts_list
+        ):
+            for pts in pts_list:
+                yield self.cache[pts]
+
+        # If not all, we ignore the cache because we
+        # need to decode and they are all consecutive
         start = t_to_pts(start, self.time_base)
         end = (
             t_to_pts(end, self.time_base)
@@ -206,6 +273,9 @@ class VideoFrameCache:
                 if frame.pts is None:
                     continue
 
+                # We store all the frames in cache
+                self._store_frame_in_cache(frame)
+
                 if frame.pts < start:
                     continue
 

yta_video_opengl/tests.py

@@ -582,6 +582,17 @@ def video_modified_stored():
     from yta_video_opengl.classes import WavingFrame, BreathingFrame, HandheldFrame, OrbitingFrame, RotatingInCenterFrame, StrangeTvFrame, GlitchRgbFrame, WavingNode
     from yta_video_opengl.utils import texture_to_frame, frame_to_texture
     from yta_video_opengl.video import Video
+    from yta_video_opengl.complete.timeline import Timeline
+
+    video = Video(VIDEO_PATH, 0.25, 0.75)
+    timeline = Timeline()
+    timeline.add_video(Video(VIDEO_PATH, 0.25, 0.75), 0.5)
+    # This is successfully raising an exception
+    #timeline.add_video(Video(VIDEO_PATH, 0.25, 0.75), 0.6)
+    timeline.add_video(Video(VIDEO_PATH, 0.25, 0.75), 1.5)
+    timeline.render(OUTPUT_PATH)
+
+    return
 
     Video(VIDEO_PATH, 0.25, 0.75).save_as(OUTPUT_PATH)
 

yta_video_opengl/utils.py

@@ -332,6 +332,17 @@ def pts_to_index(
     """
     return int(round(pts_to_t(pts, stream_time_base) * fps))
 
+def index_to_pts(
+    index: int,
+    stream_time_base: 'Fraction',
+    fps: float
+) -> int:
+    """
+    Transform a frame index into a 'pts' packet
+    timestamp.
+    """
+    return int(index / fps / stream_time_base)
+
 def pts_to_t(
     pts: int,
     stream_time_base: 'Fraction'

yta_video_opengl/video.py

@@ -22,6 +22,9 @@ class Video:
         """
         The start packet time stamp (pts), needed 
         to optimize the packet iteration process.
+
+        This timestamp is used to read the video
+        file source.
         """
         return int(self.start / self.reader.time_base)
     
@@ -32,6 +35,9 @@ class Video:
         """
         The end packet time stamp (pts), needed to
         optimize the packet iteration process.
+
+        This timestamp is used to read the video
+        file source.
         """
         return (
             int(self.end / self.reader.time_base)
@@ -73,6 +79,15 @@ class Video:
         The duration of the video.
         """
         return self.end - self.start
+    
+    @property
+    def number_of_frames(
+        self
+    ):
+        """
+        The number of frames of the video.
+        """
+        return self.reader.number_of_frames
 
     @property
     def frames(
@@ -84,6 +99,10 @@ class Video:
         'start' and 'end' parameters provided when
         instantiating it.
 
+        The iterator will iterate first over the
+        video frames, and once finished over the
+        audio frames.
+
         This method returns a tuple of 3 elements:
         - `frame` as a `VideoFrame` instance
         - `t` as the frame time moment
@@ -95,17 +114,6 @@ class Video:
         for frame in self.reader.get_audio_frames(self.start, self.end):
             yield frame
 
-        # for frame in iterate_stream_frames_demuxing(
-        #     container = self.reader.container,
-        #     video_stream = self.reader.video_stream,
-        #     audio_stream = self.reader.audio_stream,
-        #     video_start_pts = self.start_pts,
-        #     video_end_pts = self.end_pts,
-        #     audio_start_pts = self.audio_start_pts,
-        #     audio_end_pts = self.audio_end_pts
-        # ):
-        #     yield frame
-
     def __init__(
         self,
         filename: str,
@@ -143,25 +151,46 @@ class Video:
         filename: str
     ) -> 'Video':
         writer =  VideoWriter(filename)
-        #writer.set_video_stream(self.reader.video_stream.codec.name, self.reader.fps, self.reader.size, PIXEL_FORMAT)
         writer.set_video_stream_from_template(self.reader.video_stream)
         writer.set_audio_stream_from_template(self.reader.audio_stream)
 
-        # TODO: I need to process the audio also, so
-        # build a method that do the same but for 
-        # both streams at the same time
+        from yta_video_opengl.nodes.audio import VolumeAudioNode
+        # Audio from 0 to 1
+        # TODO: This effect 'fn' is shitty
+        def fade_in_fn(t, index, start=0.5, end=1.0):
+            if t < start or t > end:
+                # fuera de la franja: no tocar nada → volumen original (1.0)
+                progress = 1.0
+            else:
+                # dentro de la franja: interpolar linealmente entre 0 → 1
+                progress = (t - start) / (end - start)
+
+            return progress
+
+        #fade_in = SetVolumeAudioNode(lambda t, i: min(1, t / self.duration))
+        fade_in = VolumeAudioNode(lambda t, i: fade_in_fn(t, i, 0.5, 1.0))
+
         for frame, t, index in self.frames:
             if PythonValidator.is_instance_of(frame, 'VideoFrame'):
                 print(f'Saving video frame {str(index)}, with t = {str(t)}')
+
+                # TODO: Process any video frame change
+
                 writer.mux_video_frame(
                     frame = frame
                 )
             else:
                 print(f'Saving audio frame {str(index)} ({str(round(float(t * self.reader.fps), 2))}), with t = {str(t)}')
+
+                # TODO: Process any audio frame change
+                # Test setting audio
+                frame = fade_in.process(frame, t)
+
                 writer.mux_audio_frame(
                     frame = frame
                 )
 
+        # Flush the remaining frames to write
         writer.mux_audio_frame(None)
         writer.mux_video_frame(None)
 

yta_video_opengl/writer.py

@@ -59,6 +59,9 @@ class VideoWriter:
             options = options
         )
 
+        # We need to force this or it will not work
+        self.video_stream.time_base = Fraction(1, int(fps))
+
         if size is not None:
             self.video_stream.width = size[0]
             self.video_stream.height = size[1]
@@ -190,6 +193,7 @@ class VideoWriter:
                 # TODO: What strategy should we adopt with
                 # the packets that cannot be handled
                 # properly (?)
+                print('Invalid packet')
                 print(packet)
                 pass
 

{yta_video_opengl-0.0.9.dist-info → yta_video_opengl-0.0.11.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: yta-video-opengl
-Version: 0.0.9
+Version: 0.0.11
 Summary: Youtube Autonomous Video OpenGL Module
 Author: danialcala94
 Author-email: danielalcalavalera@gmail.com