Various fixes and improvements (#14492)

* Refactor preprocessing of images

* Cleanup preprocessing

* Improve naming and handling of embeddings

* Handle invalid intel json

* remove unused

* Use enum for model types

* Formatting

frigate/embeddings/embeddings.py

@@ -1,13 +1,11 @@
 """SQLite-vec embeddings database."""
 
 import base64
-import io
 import logging
 import os
 import time
 
 from numpy import ndarray
-from PIL import Image
 from playhouse.shortcuts import model_to_dict
 
 from frigate.comms.inter_process import InterProcessRequestor
@@ -22,7 +20,7 @@ from frigate.models import Event
 from frigate.types import ModelStatusTypesEnum
 from frigate.util.builtin import serialize
 
-from .functions.onnx import GenericONNXEmbedding
+from .functions.onnx import GenericONNXEmbedding, ModelTypeEnum
 
 logger = logging.getLogger(__name__)
 
@@ -97,7 +95,7 @@ class Embeddings:
                 "text_model_fp16.onnx": "https://huggingface.co/jinaai/jina-clip-v1/resolve/main/onnx/text_model_fp16.onnx",
             },
             model_size=config.model_size,
-            model_type="text",
+            model_type=ModelTypeEnum.text,
             requestor=self.requestor,
             device="CPU",
         )
@@ -118,83 +116,102 @@ class Embeddings:
             model_file=model_file,
             download_urls=download_urls,
             model_size=config.model_size,
-            model_type="vision",
+            model_type=ModelTypeEnum.vision,
             requestor=self.requestor,
             device="GPU" if config.model_size == "large" else "CPU",
         )
 
-    def upsert_thumbnail(self, event_id: str, thumbnail: bytes) -> ndarray:
+    def embed_thumbnail(
-        # Convert thumbnail bytes to PIL Image
+        self, event_id: str, thumbnail: bytes, upsert: bool = True
-        image = Image.open(io.BytesIO(thumbnail)).convert("RGB")
+    ) -> ndarray:
-        embedding = self.vision_embedding([image])[0]
+        """Embed thumbnail and optionally insert into DB.
 
-        self.db.execute_sql(
+        @param: event_id in Events DB
-            """
+        @param: thumbnail bytes in jpg format
-            INSERT OR REPLACE INTO vec_thumbnails(id, thumbnail_embedding)
+        @param: upsert If embedding should be upserted into vec DB
-            VALUES(?, ?)
+        """
-            """,
+        # Convert thumbnail bytes to PIL Image
-            (event_id, serialize(embedding)),
+        embedding = self.vision_embedding([thumbnail])[0]
-        )
+
+        if upsert:
+            self.db.execute_sql(
+                """
+                INSERT OR REPLACE INTO vec_thumbnails(id, thumbnail_embedding)
+                VALUES(?, ?)
+                """,
+                (event_id, serialize(embedding)),
+            )
 
         return embedding
 
-    def batch_upsert_thumbnail(self, event_thumbs: dict[str, bytes]) -> list[ndarray]:
+    def batch_embed_thumbnail(
-        images = [
+        self, event_thumbs: dict[str, bytes], upsert: bool = True
-            Image.open(io.BytesIO(thumb)).convert("RGB")
+    ) -> list[ndarray]:
-            for thumb in event_thumbs.values()
+        """Embed thumbnails and optionally insert into DB.
-        ]
+
+        @param: event_thumbs Map of Event IDs in DB to thumbnail bytes in jpg format
+        @param: upsert If embedding should be upserted into vec DB
+        """
         ids = list(event_thumbs.keys())
-        embeddings = self.vision_embedding(images)
+        embeddings = self.vision_embedding(list(event_thumbs.values()))
 
-        items = []
+        if upsert:
+            items = []
 
-        for i in range(len(ids)):
+            for i in range(len(ids)):
-            items.append(ids[i])
+                items.append(ids[i])
-            items.append(serialize(embeddings[i]))
+                items.append(serialize(embeddings[i]))
+
+            self.db.execute_sql(
+                """
+                INSERT OR REPLACE INTO vec_thumbnails(id, thumbnail_embedding)
+                VALUES {}
+                """.format(", ".join(["(?, ?)"] * len(ids))),
+                items,
+            )
 
-        self.db.execute_sql(
-            """
-            INSERT OR REPLACE INTO vec_thumbnails(id, thumbnail_embedding)
-            VALUES {}
-            """.format(", ".join(["(?, ?)"] * len(ids))),
-            items,
-        )
         return embeddings
 
-    def upsert_description(self, event_id: str, description: str) -> ndarray:
+    def embed_description(
+        self, event_id: str, description: str, upsert: bool = True
+    ) -> ndarray:
         embedding = self.text_embedding([description])[0]
-        self.db.execute_sql(
+
-            """
+        if upsert:
-            INSERT OR REPLACE INTO vec_descriptions(id, description_embedding)
+            self.db.execute_sql(
-            VALUES(?, ?)
+                """
-            """,
+                INSERT OR REPLACE INTO vec_descriptions(id, description_embedding)
-            (event_id, serialize(embedding)),
+                VALUES(?, ?)
-        )
+                """,
+                (event_id, serialize(embedding)),
+            )
 
         return embedding
 
-    def batch_upsert_description(self, event_descriptions: dict[str, str]) -> ndarray:
+    def batch_embed_description(
+        self, event_descriptions: dict[str, str], upsert: bool = True
+    ) -> ndarray:
         # upsert embeddings one by one to avoid token limit
         embeddings = []
 
         for desc in event_descriptions.values():
             embeddings.append(self.text_embedding([desc])[0])
 
-        ids = list(event_descriptions.keys())
+        if upsert:
+            ids = list(event_descriptions.keys())
+            items = []
 
-        items = []
+            for i in range(len(ids)):
+                items.append(ids[i])
+                items.append(serialize(embeddings[i]))
 
-        for i in range(len(ids)):
+            self.db.execute_sql(
-            items.append(ids[i])
+                """
-            items.append(serialize(embeddings[i]))
+                INSERT OR REPLACE INTO vec_descriptions(id, description_embedding)
-
+                VALUES {}
-        self.db.execute_sql(
+                """.format(", ".join(["(?, ?)"] * len(ids))),
-            """
+                items,
-            INSERT OR REPLACE INTO vec_descriptions(id, description_embedding)
+            )
-            VALUES {}
-            """.format(", ".join(["(?, ?)"] * len(ids))),
-            items,
-        )
 
         return embeddings
 
@@ -261,10 +278,10 @@ class Embeddings:
                 totals["processed_objects"] += 1
 
             # run batch embedding
-            self.batch_upsert_thumbnail(batch_thumbs)
+            self.batch_embed_thumbnail(batch_thumbs)
 
             if batch_descs:
-                self.batch_upsert_description(batch_descs)
+                self.batch_embed_description(batch_descs)
 
             # report progress every batch so we don't spam the logs
             progress = (totals["processed_objects"] / total_events) * 100

frigate/embeddings/functions/onnx.py

@@ -1,6 +1,7 @@
 import logging
 import os
 import warnings
+from enum import Enum
 from io import BytesIO
 from typing import Dict, List, Optional, Union
 
@@ -31,6 +32,12 @@ disable_progress_bar()
 logger = logging.getLogger(__name__)
 
 
+class ModelTypeEnum(str, Enum):
+    face = "face"
+    vision = "vision"
+    text = "text"
+
+
 class GenericONNXEmbedding:
     """Generic embedding function for ONNX models (text and vision)."""
 
@@ -88,7 +95,10 @@ class GenericONNXEmbedding:
             file_name = os.path.basename(path)
             if file_name in self.download_urls:
                 ModelDownloader.download_from_url(self.download_urls[file_name], path)
-            elif file_name == self.tokenizer_file and self.model_type == "text":
+            elif (
+                file_name == self.tokenizer_file
+                and self.model_type == ModelTypeEnum.text
+            ):
                 if not os.path.exists(path + "/" + self.model_name):
                     logger.info(f"Downloading {self.model_name} tokenizer")
                 tokenizer = AutoTokenizer.from_pretrained(
@@ -119,7 +129,7 @@ class GenericONNXEmbedding:
         if self.runner is None:
             if self.downloader:
                 self.downloader.wait_for_download()
-            if self.model_type == "text":
+            if self.model_type == ModelTypeEnum.text:
                 self.tokenizer = self._load_tokenizer()
             else:
                 self.feature_extractor = self._load_feature_extractor()
@@ -143,11 +153,35 @@ class GenericONNXEmbedding:
             f"{MODEL_CACHE_DIR}/{self.model_name}",
         )
 
+    def _preprocess_inputs(self, raw_inputs: any) -> any:
+        if self.model_type == ModelTypeEnum.text:
+            max_length = max(len(self.tokenizer.encode(text)) for text in raw_inputs)
+            return [
+                self.tokenizer(
+                    text,
+                    padding="max_length",
+                    truncation=True,
+                    max_length=max_length,
+                    return_tensors="np",
+                )
+                for text in raw_inputs
+            ]
+        elif self.model_type == ModelTypeEnum.vision:
+            processed_images = [self._process_image(img) for img in raw_inputs]
+            return [
+                self.feature_extractor(images=image, return_tensors="np")
+                for image in processed_images
+            ]
+        else:
+            raise ValueError(f"Unable to preprocess inputs for {self.model_type}")
+
     def _process_image(self, image):
         if isinstance(image, str):
             if image.startswith("http"):
                 response = requests.get(image)
                 image = Image.open(BytesIO(response.content)).convert("RGB")
+        elif isinstance(image, bytes):
+            image = Image.open(BytesIO(image)).convert("RGB")
 
         return image
 
@@ -163,25 +197,7 @@ class GenericONNXEmbedding:
             )
             return []
 
-        if self.model_type == "text":
+        processed_inputs = self._preprocess_inputs(inputs)
-            max_length = max(len(self.tokenizer.encode(text)) for text in inputs)
-            processed_inputs = [
-                self.tokenizer(
-                    text,
-                    padding="max_length",
-                    truncation=True,
-                    max_length=max_length,
-                    return_tensors="np",
-                )
-                for text in inputs
-            ]
-        else:
-            processed_images = [self._process_image(img) for img in inputs]
-            processed_inputs = [
-                self.feature_extractor(images=image, return_tensors="np")
-                for image in processed_images
-            ]
-
         input_names = self.runner.get_input_names()
         onnx_inputs = {name: [] for name in input_names}
         input: dict[str, any]

frigate/embeddings/maintainer.py

@@ -86,7 +86,7 @@ class EmbeddingMaintainer(threading.Thread):
             try:
                 if topic == EmbeddingsRequestEnum.embed_description.value:
                     return serialize(
-                        self.embeddings.upsert_description(
+                        self.embeddings.embed_description(
                             data["id"], data["description"]
                         ),
                         pack=False,
@@ -94,7 +94,7 @@ class EmbeddingMaintainer(threading.Thread):
                 elif topic == EmbeddingsRequestEnum.embed_thumbnail.value:
                     thumbnail = base64.b64decode(data["thumbnail"])
                     return serialize(
-                        self.embeddings.upsert_thumbnail(data["id"], thumbnail),
+                        self.embeddings.embed_thumbnail(data["id"], thumbnail),
                         pack=False,
                     )
                 elif topic == EmbeddingsRequestEnum.generate_search.value:
@@ -270,7 +270,7 @@ class EmbeddingMaintainer(threading.Thread):
 
     def _embed_thumbnail(self, event_id: str, thumbnail: bytes) -> None:
         """Embed the thumbnail for an event."""
-        self.embeddings.upsert_thumbnail(event_id, thumbnail)
+        self.embeddings.embed_thumbnail(event_id, thumbnail)
 
     def _embed_description(self, event: Event, thumbnails: list[bytes]) -> None:
         """Embed the description for an event."""
@@ -290,8 +290,8 @@ class EmbeddingMaintainer(threading.Thread):
             {"id": event.id, "description": description},
         )
 
-        # Encode the description
+        # Embed the description
-        self.embeddings.upsert_description(event.id, description)
+        self.embeddings.embed_description(event.id, description)
 
         logger.debug(
             "Generated description for %s (%d images): %s",

frigate/util/services.py

@@ -279,10 +279,27 @@ def get_intel_gpu_stats() -> dict[str, str]:
         logger.error(f"Unable to poll intel GPU stats: {p.stderr}")
         return None
     else:
+        output = "".join(p.stdout.split())
+
         try:
-            data = json.loads(f'[{"".join(p.stdout.split())}]')
+            data = json.loads(f"[{output}]")
         except json.JSONDecodeError:
-            return {"gpu": "-%", "mem": "-%"}
+            data = None
+
+            # json is incomplete, remove characters until we get to valid json
+            while True:
+                while output and output[-1] != "}":
+                    output = output[:-1]
+
+                if not output:
+                    return {"gpu": "", "mem": ""}
+
+                try:
+                    data = json.loads(f"[{output}]")
+                    break
+                except json.JSONDecodeError:
+                    output = output[:-1]
+                    continue
 
         results: dict[str, str] = {}
         render = {"global": []}