Merge pull request #115 from daily-co/add-vision-and-moondream-service

add vision and moondream service

README.md

@@ -39,6 +39,8 @@ Currently implemented services:
 - Transport
   - Daily
   - Local (in progress, intended as a quick start example service)
+- Vision
+  - Moondream
 
 If you'd like to [implement a service]((https://github.com/daily-co/daily-ai-sdk/tree/main/src/dailyai/services)), we welcome PRs! Our goal is to support lots of services in all of the above categories, plus new categories (like real-time video) as they emerge.
 
@@ -63,7 +65,7 @@ pip install "dailyai[option,...]"
 
 Your project may or may not need these, so they're made available as optional requirements. Here is a list:
 
-- **AI services**: `anthropic`, `azure`, `fal`, `openai`, `playht`, `silero`, `whisper`
+- **AI services**: `anthropic`, `azure`, `fal`, `moondream`, `openai`, `playht`, `silero`, `whisper`
 - **Transports**: `daily`, `local`, `websocket`
 
 ## Code examples

examples/foundational/12-describe-video.py

@@ -0,0 +1,84 @@
+import asyncio
+import aiohttp
+import logging
+import os
+
+from typing import AsyncGenerator
+
+from dailyai.pipeline.aggregators import FrameProcessor, UserResponseAggregator, VisionImageFrameAggregator
+
+from dailyai.pipeline.frames import Frame, TextFrame, UserImageRequestFrame
+from dailyai.pipeline.pipeline import Pipeline
+from dailyai.services.elevenlabs_ai_service import ElevenLabsTTSService
+from dailyai.services.moondream_ai_service import MoondreamService
+from dailyai.transports.daily_transport import DailyTransport
+
+from runner import configure
+
+from dotenv import load_dotenv
+load_dotenv(override=True)
+
+logging.basicConfig(format=f"%(levelno)s %(asctime)s %(message)s")
+logger = logging.getLogger("dailyai")
+logger.setLevel(logging.DEBUG)
+
+
+class UserImageRequester(FrameProcessor):
+    participant_id: str
+
+    def set_participant_id(self, participant_id: str):
+        self.participant_id = participant_id
+
+    async def process_frame(self, frame: Frame) -> AsyncGenerator[Frame, None]:
+        if self.participant_id and isinstance(frame, TextFrame):
+            yield UserImageRequestFrame(self.participant_id)
+        yield frame
+
+
+async def main(room_url: str, token):
+    async with aiohttp.ClientSession() as session:
+        transport = DailyTransport(
+            room_url,
+            token,
+            "Describe participant video",
+            duration_minutes=5,
+            mic_enabled=True,
+            mic_sample_rate=16000,
+            vad_enabled=True,
+            start_transcription=True,
+            video_rendering_enabled=True
+        )
+
+        tts = ElevenLabsTTSService(
+            aiohttp_session=session,
+            api_key=os.getenv("ELEVENLABS_API_KEY"),
+            voice_id=os.getenv("ELEVENLABS_VOICE_ID"),
+        )
+
+        user_response = UserResponseAggregator()
+
+        image_requester = UserImageRequester()
+
+        vision_aggregator = VisionImageFrameAggregator()
+
+        moondream = MoondreamService()
+
+        tts = ElevenLabsTTSService(
+            aiohttp_session=session,
+            api_key=os.getenv("ELEVENLABS_API_KEY"),
+            voice_id=os.getenv("ELEVENLABS_VOICE_ID"),
+        )
+
+        @transport.event_handler("on_first_other_participant_joined")
+        async def on_first_other_participant_joined(transport, participant):
+            await transport.say("Hi there! Feel free to ask me what I see.", tts)
+            transport.render_participant_video(participant["id"], framerate=0)
+            image_requester.set_participant_id(participant["id"])
+
+        pipeline = Pipeline([user_response, image_requester, vision_aggregator, moondream, tts])
+
+        await transport.run(pipeline)
+
+if __name__ == "__main__":
+    (url, token) = configure()
+    asyncio.run(main(url, token))

examples/foundational/14-render-remote-participant.py

@@ -1,5 +1,4 @@
 import asyncio
-import io
 import logging
 
 from typing import AsyncGenerator

examples/foundational/14a-local-render-remote-participant.py

@@ -1,5 +1,4 @@
 import asyncio
-import io
 import logging
 import tkinter as tk
 

linux-py3.10-requirements.txt

@@ -50,7 +50,7 @@ cryptography==42.0.5
     # via pyjwt
 ctranslate2==4.1.0
     # via faster-whisper
-daily-python==0.7.2
+daily-python==0.7.3
     # via dailyai (pyproject.toml)
 deprecated==1.2.14
     # via opentelemetry-api
@@ -62,6 +62,8 @@ distro==1.9.0
     # via
     #   anthropic
     #   openai
+einops==0.7.0
+    # via dailyai (pyproject.toml)
 exceptiongroup==1.2.0
     # via anyio
 fal==0.12.7
@@ -70,11 +72,12 @@ fastapi==0.99.1
     # via fal
 faster-whisper==1.0.1
     # via dailyai (pyproject.toml)
-filelock==3.13.3
+filelock==3.13.4
     # via
     #   huggingface-hub
     #   pyht
     #   torch
+    #   transformers
     #   triton
     #   virtualenv
 flask==3.0.3
@@ -114,7 +117,9 @@ httpx==0.27.0
 huggingface-hub==0.22.2
     # via
     #   faster-whisper
+    #   timm
     #   tokenizers
+    #   transformers
 humanfriendly==10.0
     # via coloredlogs
 idna==3.6
@@ -160,6 +165,8 @@ numpy==1.26.4
     #   ctranslate2
     #   dailyai (pyproject.toml)
     #   onnxruntime
+    #   torchvision
+    #   transformers
 nvidia-cublas-cu12==12.1.3.1
     # via
     #   nvidia-cudnn-cu12
@@ -208,12 +215,14 @@ packaging==24.0
     #   fal
     #   huggingface-hub
     #   onnxruntime
+    #   transformers
 pathspec==0.11.2
     # via fal
 pillow==10.2.0
     # via
     #   dailyai (pyproject.toml)
     #   fal
+    #   torchvision
 platformdirs==4.2.0
     # via
     #   isolate
@@ -251,16 +260,25 @@ pyyaml==6.0.1
     #   ctranslate2
     #   huggingface-hub
     #   isolate
+    #   timm
+    #   transformers
+regex==2023.12.25
+    # via transformers
 requests==2.31.0
     # via
     #   huggingface-hub
     #   pyht
+    #   transformers
 rich==13.7.1
     # via
     #   fal
     #   rich-click
 rich-click==1.7.4
     # via fal
+safetensors==0.4.2
+    # via
+    #   timm
+    #   transformers
 six==1.16.0
     # via python-dateutil
 sniffio==1.3.1
@@ -279,20 +297,30 @@ sympy==1.12
     #   torch
 tblib==3.0.0
     # via isolate
+timm==0.9.16
+    # via dailyai (pyproject.toml)
 tokenizers==0.15.2
     # via
     #   anthropic
     #   faster-whisper
+    #   transformers
 torch==2.2.2
     # via
     #   dailyai (pyproject.toml)
+    #   timm
     #   torchaudio
+    #   torchvision
 torchaudio==2.2.2
     # via dailyai (pyproject.toml)
+torchvision==0.17.2
+    # via timm
 tqdm==4.66.2
     # via
     #   huggingface-hub
     #   openai
+    #   transformers
+transformers==4.39.3
+    # via dailyai (pyproject.toml)
 triton==2.2.0
     # via torch
 types-python-dateutil==2.9.0.20240316

macos-py3.10-requirements.txt

@@ -50,7 +50,7 @@ cryptography==42.0.5
     # via pyjwt
 ctranslate2==4.1.0
     # via faster-whisper
-daily-python==0.7.2
+daily-python==0.7.3
     # via dailyai (pyproject.toml)
 deprecated==1.2.14
     # via opentelemetry-api
@@ -62,6 +62,8 @@ distro==1.9.0
     # via
     #   anthropic
     #   openai
+einops==0.7.0
+    # via dailyai (pyproject.toml)
 exceptiongroup==1.2.0
     # via anyio
 fal==0.12.7
@@ -70,11 +72,12 @@ fastapi==0.99.1
     # via fal
 faster-whisper==1.0.1
     # via dailyai (pyproject.toml)
-filelock==3.13.3
+filelock==3.13.4
     # via
     #   huggingface-hub
     #   pyht
     #   torch
+    #   transformers
     #   virtualenv
 flask==3.0.3
     # via
@@ -113,7 +116,9 @@ httpx==0.27.0
 huggingface-hub==0.22.2
     # via
     #   faster-whisper
+    #   timm
     #   tokenizers
+    #   transformers
 humanfriendly==10.0
     # via coloredlogs
 idna==3.6
@@ -159,6 +164,8 @@ numpy==1.26.4
     #   ctranslate2
     #   dailyai (pyproject.toml)
     #   onnxruntime
+    #   torchvision
+    #   transformers
 onnxruntime==1.17.1
     # via faster-whisper
 openai==1.14.3
@@ -176,12 +183,14 @@ packaging==24.0
     #   fal
     #   huggingface-hub
     #   onnxruntime
+    #   transformers
 pathspec==0.11.2
     # via fal
 pillow==10.2.0
     # via
     #   dailyai (pyproject.toml)
     #   fal
+    #   torchvision
 platformdirs==4.2.0
     # via
     #   isolate
@@ -219,16 +228,25 @@ pyyaml==6.0.1
     #   ctranslate2
     #   huggingface-hub
     #   isolate
+    #   timm
+    #   transformers
+regex==2023.12.25
+    # via transformers
 requests==2.31.0
     # via
     #   huggingface-hub
     #   pyht
+    #   transformers
 rich==13.7.1
     # via
     #   fal
     #   rich-click
 rich-click==1.7.4
     # via fal
+safetensors==0.4.2
+    # via
+    #   timm
+    #   transformers
 six==1.16.0
     # via python-dateutil
 sniffio==1.3.1
@@ -247,20 +265,30 @@ sympy==1.12
     #   torch
 tblib==3.0.0
     # via isolate
+timm==0.9.16
+    # via dailyai (pyproject.toml)
 tokenizers==0.15.2
     # via
     #   anthropic
     #   faster-whisper
+    #   transformers
 torch==2.2.2
     # via
     #   dailyai (pyproject.toml)
+    #   timm
     #   torchaudio
+    #   torchvision
 torchaudio==2.2.2
     # via dailyai (pyproject.toml)
+torchvision==0.17.2
+    # via timm
 tqdm==4.66.2
     # via
     #   huggingface-hub
     #   openai
+    #   transformers
+transformers==4.39.3
+    # via dailyai (pyproject.toml)
 types-python-dateutil==2.9.0.20240316
     # via fal
 typing-extensions==4.10.0

pyproject.toml

@@ -37,6 +37,7 @@ daily = [ "daily-python~=0.7.0" ]
 examples = [ "python-dotenv~=1.0.0", "flask~=3.0.0", "flask_cors~=4.0.0" ]
 fal = [ "fal~=0.12.0" ]
 local = [ "pyaudio~=0.2.0" ]
+moondream = [ "einops~=0.7.0", "timm~=0.9.0", "transformers~=4.39.0" ]
 openai = [ "openai~=1.14.0" ]
 playht = [ "pyht~=0.0.26" ]
 silero = [ "torch~=2.2.0", "torchaudio~=2.2.0" ]

src/dailyai/pipeline/aggregators.py

@@ -7,6 +7,7 @@ from dailyai.pipeline.frames import (
     EndFrame,
     EndPipeFrame,
     Frame,
+    ImageFrame,
     LLMMessagesFrame,
     LLMResponseEndFrame,
     LLMResponseStartFrame,
@@ -14,6 +15,7 @@ from dailyai.pipeline.frames import (
     TranscriptionFrame,
     UserStartedSpeakingFrame,
     UserStoppedSpeakingFrame,
+    VisionImageFrame,
 )
 from dailyai.pipeline.pipeline import Pipeline
 from dailyai.services.ai_services import AIService
@@ -463,3 +465,37 @@ class GatedAggregator(FrameProcessor):
             self.accumulator = []
         else:
             self.accumulator.append(frame)
+
+
+class VisionImageFrameAggregator(FrameProcessor):
+    """This aggregator waits for a consecutive TextFrame and an
+    ImageFrame. After the ImageFrame arrives it will output a VisionImageFrame.
+
+    >>> from dailyai.pipeline.frames import ImageFrame
+
+    >>> async def print_frames(aggregator, frame):
+    ...     async for frame in aggregator.process_frame(frame):
+    ...         print(frame)
+
+    >>> aggregator = VisionImageFrameAggregator()
+    >>> asyncio.run(print_frames(aggregator, TextFrame("What do you see?")))
+    >>> asyncio.run(print_frames(aggregator, ImageFrame(image=bytes([]), size=(0, 0))))
+    VisionImageFrame, text: What do you see?, image size: 0x0, buffer size: 0 B
+
+    """
+
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+        self._describe_text = None
+
+    async def process_frame(self, frame: Frame) -> AsyncGenerator[Frame, None]:
+        if isinstance(frame, TextFrame):
+            self._describe_text = frame.text
+        elif isinstance(frame, ImageFrame):
+            if self._describe_text:
+                yield VisionImageFrame(self._describe_text, frame.image, frame.size)
+                self._describe_text = None
+            else:
+                yield frame
+        else:
+            yield frame

src/dailyai/pipeline/frames.py

@@ -79,8 +79,10 @@ class ImageFrame(Frame):
 
 @dataclass()
 class URLImageFrame(ImageFrame):
-    """An image. Will be shown by the transport if the transport's camera is
-    enabled."""
+    """An image with an associated URL. Will be shown by the transport if the
+    transport's camera is enabled.
+
+    """
     url: str | None
 
     def __init__(self, url, image, size):
@@ -91,6 +93,22 @@ class URLImageFrame(ImageFrame):
         return f"{self.__class__.__name__}, url: {self.url}, image size: {self.size[0]}x{self.size[1]}, buffer size: {len(self.image)} B"
 
 
+@dataclass()
+class VisionImageFrame(ImageFrame):
+    """An image with an associated text to ask for a description of it. Will be shown by the
+    transport if the transport's camera is enabled.
+
+    """
+    text: str | None
+
+    def __init__(self, text, image, size):
+        super().__init__(image, size)
+        self.text = text
+
+    def __str__(self):
+        return f"{self.__class__.__name__}, text: {self.text}, image size: {self.size[0]}x{self.size[1]}, buffer size: {len(self.image)} B"
+
+
 @dataclass()
 class UserImageFrame(ImageFrame):
     """An image associated to a user. Will be shown by the transport if the transport's camera is
@@ -105,6 +123,15 @@ class UserImageFrame(ImageFrame):
         return f"{self.__class__.__name__}, user: {self.user_id}, image size: {self.size[0]}x{self.size[1]}, buffer size: {len(self.image)} B"
 
 
+@dataclass()
+class UserImageRequestFrame(Frame):
+    """A frame user to request an image from the given user."""
+    user_id: str
+
+    def __str__(self):
+        return f"{self.__class__.__name__}, user: {self.user_id}"
+
+
 @dataclass()
 class SpriteFrame(Frame):
     """An animated sprite. Will be shown by the transport if the transport's
@@ -172,10 +199,10 @@ class ReceivedAppMessageFrame(Frame):
 @dataclass()
 class SendAppMessageFrame(Frame):
     message: Any
-    participantId: str | None
+    participant_id: str | None
 
     def __str__(self):
-        return f"SendAppMessageFrame: participantId: {self.participantId}, message: {self.message}"
+        return f"SendAppMessageFrame: participant: {self.participant_id}, message: {self.message}"
 
 
 class UserStartedSpeakingFrame(Frame):

src/dailyai/services/ai_services.py

@@ -15,6 +15,7 @@ from dailyai.pipeline.frames import (
     TextFrame,
     TranscriptionFrame,
     URLImageFrame,
+    VisionImageFrame,
 )
 
 from abc import abstractmethod
@@ -100,6 +101,25 @@ class ImageGenService(AIService):
         yield URLImageFrame(url, image_data, image_size)
 
 
+class VisionService(AIService):
+    """VisionService is a base class for vision services."""
+
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+        self._describe_text = None
+
+    @abstractmethod
+    async def run_vision(self, frame: VisionImageFrame) -> str:
+        pass
+
+    async def process_frame(self, frame: Frame) -> AsyncGenerator[Frame, None]:
+        if isinstance(frame, VisionImageFrame):
+            description = await self.run_vision(frame)
+            yield TextFrame(description)
+        else:
+            yield frame
+
+
 class STTService(AIService):
     """STTService is a base class for speech-to-text services."""
 

src/dailyai/services/moondream_ai_service.py

@@ -0,0 +1,52 @@
+from dailyai.pipeline.frames import ImageFrame, VisionImageFrame
+from dailyai.services.ai_services import VisionService
+
+from PIL import Image
+
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+import torch
+
+
+def detect_device():
+    """
+    Detects the appropriate device to run on, and return the device and dtype.
+    """
+    if torch.cuda.is_available():
+        return torch.device("cuda"), torch.float16
+    elif torch.backends.mps.is_available():
+        return torch.device("mps"), torch.float16
+    else:
+        return torch.device("cpu"), torch.float32
+
+
+class MoondreamService(VisionService):
+    def __init__(
+        self,
+        model_id="vikhyatk/moondream2",
+        revision="2024-04-02",
+        device=None
+    ):
+        super().__init__()
+
+        if not device:
+            device, dtype = detect_device()
+        else:
+            device = torch.device("cpu")
+            dtype = torch.float32
+
+        self._tokenizer = AutoTokenizer.from_pretrained(model_id, revision=revision)
+
+        self._model = AutoModelForCausalLM.from_pretrained(
+            model_id, trust_remote_code=True, revision=revision
+        ).to(device=device, dtype=dtype)
+        self._model.eval()
+
+    async def run_vision(self, frame: VisionImageFrame) -> str:
+        image = Image.frombytes("RGB", (frame.size[0], frame.size[1]), frame.image)
+        image_embeds = self._model.encode_image(image)
+        description = self._model.answer_question(
+            image_embeds=image_embeds,
+            question=frame.text,
+            tokenizer=self._tokenizer)
+        return description

src/dailyai/transports/daily_transport.py

@@ -169,8 +169,12 @@ class DailyTransport(ThreadedTransport, EventHandler):
         if self._mic_enabled:
             self.mic.write_frames(frame)
 
-    def send_app_message(self, message: Any, participantId: str | None):
-        self.client.send_app_message(message, participantId)
+    def request_participant_image(self, participant_id: str):
+        if participant_id in self._video_renderers:
+            self._video_renderers[participant_id]["render_next_frame"] = True
+
+    def send_app_message(self, message: Any, participant_id: str | None):
+        self.client.send_app_message(message, participant_id)
 
     def read_audio_frames(self, desired_frame_count):
         bytes = b""
@@ -302,6 +306,7 @@ class DailyTransport(ThreadedTransport, EventHandler):
         self._video_renderers[participant_id] = {
             "framerate": framerate,
             "timestamp": 0,
+            "render_next_frame": False,
         }
         self.client.set_video_renderer(
             participant_id,
@@ -310,17 +315,28 @@ class DailyTransport(ThreadedTransport, EventHandler):
             color_format=color_format)
 
     def on_participant_video_frame(self, participant_id, video_frame):
+        if not self._loop:
+            return
+
+        render_frame = False
+
         curr_time = time.time()
-        prev_time = self._video_renderers[participant_id]["timestamp"]
-        diff_time = curr_time - prev_time
-        period = 1 / self._video_renderers[participant_id]["framerate"]
-        if diff_time > period and self._loop:
-            self._video_renderers[participant_id]["timestamp"] = curr_time
+        framerate = self._video_renderers[participant_id]["framerate"]
+
+        if framerate > 0:
+            prev_time = self._video_renderers[participant_id]["timestamp"]
+            next_time = prev_time + 1 / framerate
+            render_frame = curr_time > next_time
+        elif self._video_renderers[participant_id]["render_next_frame"]:
+            self._video_renderers[participant_id]["render_next_frame"] = False
+            render_frame = True
+
+        if render_frame:
             frame = UserImageFrame(participant_id, video_frame.buffer,
                                    (video_frame.width, video_frame.height))
-            asyncio.run_coroutine_threadsafe(
-                self.receive_queue.put(frame), self._loop
-            )
+            asyncio.run_coroutine_threadsafe(self.receive_queue.put(frame), self._loop)
+
+        self._video_renderers[participant_id]["timestamp"] = curr_time
 
     def on_error(self, error):
         self._logger.error(f"on_error: {error}")

src/dailyai/transports/threaded_transport.py

@@ -20,6 +20,7 @@ from dailyai.pipeline.frames import (
     SpriteFrame,
     StartFrame,
     TextFrame,
+    UserImageRequestFrame,
     UserStartedSpeakingFrame,
     UserStoppedSpeakingFrame,
 )
@@ -382,7 +383,11 @@ class ThreadedTransport(AbstractTransport):
     def _set_images(self, images: list[bytes], start_frame=0):
         self._images = itertools.cycle(images)
 
-    def send_app_message(self, message: Any, participantId: str | None):
+    def request_participant_image(self, participant_id: str):
+        """ Child classes should override this to force an image from a user. """
+        pass
+
+    def send_app_message(self, message: Any, participant_id: str | None):
         """ Child classes should override this to send a custom message to the room. """
         pass
 
@@ -458,9 +463,10 @@ class ThreadedTransport(AbstractTransport):
                                 self._set_image(frame.image)
                             elif isinstance(frame, SpriteFrame):
                                 self._set_images(frame.images)
+                            elif isinstance(frame, UserImageRequestFrame):
+                                self.request_participant_image(frame.user_id)
                             elif isinstance(frame, SendAppMessageFrame):
-                                self.send_app_message(
-                                    frame.message, frame.participantId)
+                                self.send_app_message(frame.message, frame.participant_id)
                         elif len(b):
                             self.write_frame_to_mic(bytes(b))
                             b = bytearray()