diff --git a/.cspell/frigate-dictionary.txt b/.cspell/frigate-dictionary.txt index 6e66a4704..329c41815 100644 --- a/.cspell/frigate-dictionary.txt +++ b/.cspell/frigate-dictionary.txt @@ -191,6 +191,7 @@ ONVIF openai opencv openvino +overfitting OWASP paddleocr paho diff --git a/docs/docs/configuration/audio_detectors.md b/docs/docs/configuration/audio_detectors.md index 80b0727a5..f2ff99b6b 100644 --- a/docs/docs/configuration/audio_detectors.md +++ b/docs/docs/configuration/audio_detectors.md @@ -168,6 +168,8 @@ Recorded `speech` events will always use a `whisper` model, regardless of the `m If you hear speech that’s actually important and worth saving/indexing for the future, **just press the transcribe button in Explore** on that specific `speech` event - that keeps things explicit, reliable, and under your control. + Other options are being considered for future versions of Frigate to add transcription options that support external `whisper` Docker containers. A single transcription service could then be shared by Frigate and other applications (for example, Home Assistant Voice), and run on more powerful machines when available. + 2. Why don't you save live transcription text and use that for `speech` events? There’s no guarantee that a `speech` event is even created from the exact audio that went through the transcription model. Live transcription and `speech` event creation are **separate, asynchronous processes**. Even when both are correctly configured, trying to align the **precise start and end time of a speech event** with whatever audio the model happened to be processing at that moment is unreliable. diff --git a/docs/docs/configuration/custom_classification/state_classification.md b/docs/docs/configuration/custom_classification/state_classification.md index 66d3e60ca..927fe91af 100644 --- a/docs/docs/configuration/custom_classification/state_classification.md +++ b/docs/docs/configuration/custom_classification/state_classification.md @@ -69,4 +69,6 @@ Once all images are assigned, training will begin automatically. ### Improving the Model - **Problem framing**: Keep classes visually distinct and state-focused (e.g., `open`, `closed`, `unknown`). Avoid combining object identity with state in a single model unless necessary. -- **Data collection**: Use the model’s Recent Classifications tab to gather balanced examples across times of day and weather. +- **Data collection**: Use the model's Recent Classifications tab to gather balanced examples across times of day and weather. +- **When to train**: Focus on cases where the model is entirely incorrect or flips between states when it should not. There's no need to train additional images when the model is already working consistently. +- **Selecting training images**: Images scoring below 100% due to new conditions (e.g., first snow of the year, seasonal changes) or variations (e.g., objects temporarily in view, insects at night) are good candidates for training, as they represent scenarios different from the default state. Training these lower-scoring images that differ from existing training data helps prevent overfitting. Avoid training large quantities of images that look very similar, especially if they already score 100% as this can lead to overfitting. diff --git a/docs/docs/configuration/reference.md b/docs/docs/configuration/reference.md index f8b49303f..a375086cb 100644 --- a/docs/docs/configuration/reference.md +++ b/docs/docs/configuration/reference.md @@ -710,6 +710,44 @@ audio_transcription: # List of language codes: https://github.com/openai/whisper/blob/main/whisper/tokenizer.py#L10 language: en +# Optional: Configuration for classification models +classification: + # Optional: Configuration for bird classification + bird: + # Optional: Enable bird classification (default: shown below) + enabled: False + # Optional: Minimum classification score required to be considered a match (default: shown below) + threshold: 0.9 + custom: + # Required: name of the classification model + model_name: + # Optional: Enable running the model (default: shown below) + enabled: True + # Optional: Name of classification model (default: shown below) + name: None + # Optional: Classification score threshold to change the state (default: shown below) + threshold: 0.8 + # Optional: Number of classification attempts to save in the recent classifications tab (default: shown below) + # NOTE: Defaults to 200 for object classification and 100 for state classification if not specified + save_attempts: None + # Optional: Object classification configuration + object_config: + # Required: Object types to classify + objects: [dog] + # Optional: Type of classification that is applied (default: shown below) + classification_type: sub_label + # Optional: State classification configuration + state_config: + # Required: Cameras to run classification on + cameras: + camera_name: + # Required: Crop of image frame on this camera to run classification on + crop: [0, 180, 220, 400] + # Optional: If classification should be run when motion is detected in the crop (default: shown below) + motion: False + # Optional: Interval to run classification on in seconds (default: shown below) + interval: None + # Optional: Restream configuration # Uses https://github.com/AlexxIT/go2rtc (v1.9.10) # NOTE: The default go2rtc API port (1984) must be used, diff --git a/frigate/api/event.py b/frigate/api/event.py index 8e966d98b..b8b596cde 100644 --- a/frigate/api/event.py +++ b/frigate/api/event.py @@ -1731,37 +1731,40 @@ def create_trigger_embedding( if event.data.get("type") != "object": return - if thumbnail := get_event_thumbnail_bytes(event): - cursor = context.db.execute_sql( - """ - SELECT thumbnail_embedding FROM vec_thumbnails WHERE id = ? - """, - [body.data], + # Get the thumbnail + thumbnail = get_event_thumbnail_bytes(event) + + if thumbnail is None: + return JSONResponse( + content={ + "success": False, + "message": f"Failed to get thumbnail for {body.data} for {body.type} trigger", + }, + status_code=400, ) - row = cursor.fetchone() if cursor else None + # Try to reuse existing embedding from database + cursor = context.db.execute_sql( + """ + SELECT thumbnail_embedding FROM vec_thumbnails WHERE id = ? + """, + [body.data], + ) - if row: - query_embedding = row[0] - embedding = np.frombuffer(query_embedding, dtype=np.float32) + row = cursor.fetchone() if cursor else None + + if row: + query_embedding = row[0] + embedding = np.frombuffer(query_embedding, dtype=np.float32) else: - # Extract valid thumbnail - thumbnail = get_event_thumbnail_bytes(event) - - if thumbnail is None: - return JSONResponse( - content={ - "success": False, - "message": f"Failed to get thumbnail for {body.data} for {body.type} trigger", - }, - status_code=400, - ) - + # Generate new embedding embedding = context.generate_image_embedding( body.data, (base64.b64encode(thumbnail).decode("ASCII")) ) - if not embedding: + if embedding is None or ( + isinstance(embedding, (list, np.ndarray)) and len(embedding) == 0 + ): return JSONResponse( content={ "success": False, @@ -1896,7 +1899,9 @@ def update_trigger_embedding( body.data, (base64.b64encode(thumbnail).decode("ASCII")) ) - if not embedding: + if embedding is None or ( + isinstance(embedding, (list, np.ndarray)) and len(embedding) == 0 + ): return JSONResponse( content={ "success": False, diff --git a/frigate/config/classification.py b/frigate/config/classification.py index bdcbf48f1..fb8e3de29 100644 --- a/frigate/config/classification.py +++ b/frigate/config/classification.py @@ -105,6 +105,11 @@ class CustomClassificationConfig(FrigateBaseModel): threshold: float = Field( default=0.8, title="Classification score threshold to change the state." ) + save_attempts: int | None = Field( + default=None, + title="Number of classification attempts to save in the recent classifications tab. If not specified, defaults to 200 for object classification and 100 for state classification.", + ge=0, + ) object_config: CustomClassificationObjectConfig | None = Field(default=None) state_config: CustomClassificationStateConfig | None = Field(default=None) diff --git a/frigate/data_processing/real_time/custom_classification.py b/frigate/data_processing/real_time/custom_classification.py index 179d2f43f..c8f31db76 100644 --- a/frigate/data_processing/real_time/custom_classification.py +++ b/frigate/data_processing/real_time/custom_classification.py @@ -250,6 +250,11 @@ class CustomStateClassificationProcessor(RealTimeProcessorApi): if self.interpreter is None: # When interpreter is None, always save (score is 0.0, which is < 1.0) if self._should_save_image(camera, "unknown", 0.0): + save_attempts = ( + self.model_config.save_attempts + if self.model_config.save_attempts is not None + else 100 + ) write_classification_attempt( self.train_dir, cv2.cvtColor(frame, cv2.COLOR_RGB2BGR), @@ -257,6 +262,7 @@ class CustomStateClassificationProcessor(RealTimeProcessorApi): now, "unknown", 0.0, + max_files=save_attempts, ) return @@ -277,6 +283,11 @@ class CustomStateClassificationProcessor(RealTimeProcessorApi): detected_state = self.labelmap[best_id] if self._should_save_image(camera, detected_state, score): + save_attempts = ( + self.model_config.save_attempts + if self.model_config.save_attempts is not None + else 100 + ) write_classification_attempt( self.train_dir, cv2.cvtColor(frame, cv2.COLOR_RGB2BGR), @@ -284,6 +295,7 @@ class CustomStateClassificationProcessor(RealTimeProcessorApi): now, detected_state, score, + max_files=save_attempts, ) if score < self.model_config.threshold: @@ -482,6 +494,11 @@ class CustomObjectClassificationProcessor(RealTimeProcessorApi): return if self.interpreter is None: + save_attempts = ( + self.model_config.save_attempts + if self.model_config.save_attempts is not None + else 200 + ) write_classification_attempt( self.train_dir, cv2.cvtColor(crop, cv2.COLOR_RGB2BGR), @@ -489,6 +506,7 @@ class CustomObjectClassificationProcessor(RealTimeProcessorApi): now, "unknown", 0.0, + max_files=save_attempts, ) return @@ -506,6 +524,11 @@ class CustomObjectClassificationProcessor(RealTimeProcessorApi): score = round(probs[best_id], 2) self.__update_metrics(datetime.datetime.now().timestamp() - now) + save_attempts = ( + self.model_config.save_attempts + if self.model_config.save_attempts is not None + else 200 + ) write_classification_attempt( self.train_dir, cv2.cvtColor(crop, cv2.COLOR_RGB2BGR), @@ -513,7 +536,7 @@ class CustomObjectClassificationProcessor(RealTimeProcessorApi): now, self.labelmap[best_id], score, - max_files=200, + max_files=save_attempts, ) if score < self.model_config.threshold: diff --git a/frigate/storage.py b/frigate/storage.py index ee11cf7a9..feabe06ff 100644 --- a/frigate/storage.py +++ b/frigate/storage.py @@ -5,7 +5,7 @@ import shutil import threading from pathlib import Path -from peewee import fn +from peewee import SQL, fn from frigate.config import FrigateConfig from frigate.const import RECORD_DIR @@ -44,13 +44,19 @@ class StorageMaintainer(threading.Thread): ) } - # calculate MB/hr + # calculate MB/hr from last 100 segments try: - bandwidth = round( - Recordings.select(fn.AVG(bandwidth_equation)) + # Subquery to get last 100 segments, then average their bandwidth + last_100 = ( + Recordings.select(bandwidth_equation.alias("bw")) .where(Recordings.camera == camera, Recordings.segment_size > 0) + .order_by(Recordings.start_time.desc()) .limit(100) - .scalar() + .alias("recent") + ) + + bandwidth = round( + Recordings.select(fn.AVG(SQL("bw"))).from_(last_100).scalar() * 3600, 2, ) diff --git a/frigate/util/classification.py b/frigate/util/classification.py index a74094c32..1f4213315 100644 --- a/frigate/util/classification.py +++ b/frigate/util/classification.py @@ -330,7 +330,7 @@ def collect_state_classification_examples( 1. Queries review items from specified cameras 2. Selects 100 balanced timestamps across the data 3. Extracts keyframes from recordings (cropped to specified regions) - 4. Selects 20 most visually distinct images + 4. Selects 24 most visually distinct images 5. Saves them to the dataset directory Args: @@ -660,7 +660,6 @@ def collect_object_classification_examples( Args: model_name: Name of the classification model label: Object label to collect (e.g., "person", "car") - cameras: List of camera names to collect examples from """ dataset_dir = os.path.join(CLIPS_DIR, model_name, "dataset") temp_dir = os.path.join(dataset_dir, "temp") diff --git a/web/src/components/overlay/detail/SearchDetailDialog.tsx b/web/src/components/overlay/detail/SearchDetailDialog.tsx index 4ead27218..1c46213df 100644 --- a/web/src/components/overlay/detail/SearchDetailDialog.tsx +++ b/web/src/components/overlay/detail/SearchDetailDialog.tsx @@ -498,7 +498,7 @@ export default function SearchDetailDialog({ const views = [...SEARCH_TABS]; - if (search.data.type != "object" || !search.has_clip) { + if (!search.has_clip) { const index = views.indexOf("tracking_details"); views.splice(index, 1); } @@ -548,7 +548,7 @@ export default function SearchDetailDialog({ "relative flex items-center justify-between", "w-full", // match dialog's max-width classes - "sm:max-w-xl md:max-w-4xl lg:max-w-[70%]", + "max-h-[95dvh] max-w-[85%] xl:max-w-[70%]", )} > @@ -594,8 +594,7 @@ export default function SearchDetailDialog({ ref={isDesktop ? dialogContentRef : undefined} className={cn( "scrollbar-container overflow-y-auto", - isDesktop && - "max-h-[95dvh] sm:max-w-xl md:max-w-4xl lg:max-w-[70%]", + isDesktop && "max-h-[95dvh] max-w-[85%] xl:max-w-[70%]", isMobile && "flex h-full flex-col px-4", )} onEscapeKeyDown={(event) => { diff --git a/web/src/components/overlay/detail/TrackingDetails.tsx b/web/src/components/overlay/detail/TrackingDetails.tsx index 26cba7d3a..fa2f0cb2b 100644 --- a/web/src/components/overlay/detail/TrackingDetails.tsx +++ b/web/src/components/overlay/detail/TrackingDetails.tsx @@ -622,7 +622,7 @@ export function TrackingDetails({
{isDesktop && tabs && ( @@ -900,96 +900,99 @@ function LifecycleIconRow({
{getLifecycleItemDescription(item)}
-
-
- - {t("trackingDetails.lifecycleItemDesc.header.score")} - - {score} -
-
- - {t("trackingDetails.lifecycleItemDesc.header.ratio")} - - {ratio} -
-
- - {t("trackingDetails.lifecycleItemDesc.header.area")}{" "} - {attributeAreaPx !== undefined && - attributeAreaPct !== undefined && ( - - ({getTranslatedLabel(item.data.label)}) - - )} - - {areaPx !== undefined && areaPct !== undefined ? ( - - {t("information.pixels", { ns: "common", area: areaPx })} ·{" "} - {areaPct}% + {/* Only show Score/Ratio/Area for object events, not for audio (heard) or manual API (external) events */} + {item.class_type !== "heard" && item.class_type !== "external" && ( +
+
+ + {t("trackingDetails.lifecycleItemDesc.header.score")} - ) : ( - N/A - )} -
- {attributeAreaPx !== undefined && - attributeAreaPct !== undefined && ( -
- - {t("trackingDetails.lifecycleItemDesc.header.area")} ( - {getTranslatedLabel(item.data.attribute)}) - - - {t("information.pixels", { - ns: "common", - area: attributeAreaPx, - })}{" "} - · {attributeAreaPct}% - -
- )} - - {item.data?.zones && item.data.zones.length > 0 && ( -
- {item.data.zones.map((zone, zidx) => { - const color = getZoneColor(zone)?.join(",") ?? "0,0,0"; - return ( - { - e.stopPropagation(); - setSelectedZone(zone); - }} - style={{ - borderColor: `rgba(${color}, 0.6)`, - background: `rgba(${color}, 0.08)`, - }} - > - - - {item.data?.zones_friendly_names?.[zidx]} - - - ); - })} + {score}
- )} -
+
+ + {t("trackingDetails.lifecycleItemDesc.header.ratio")} + + {ratio} +
+
+ + {t("trackingDetails.lifecycleItemDesc.header.area")}{" "} + {attributeAreaPx !== undefined && + attributeAreaPct !== undefined && ( + + ({getTranslatedLabel(item.data.label)}) + + )} + + {areaPx !== undefined && areaPct !== undefined ? ( + + {t("information.pixels", { ns: "common", area: areaPx })}{" "} + · {areaPct}% + + ) : ( + N/A + )} +
+ {attributeAreaPx !== undefined && + attributeAreaPct !== undefined && ( +
+ + {t("trackingDetails.lifecycleItemDesc.header.area")} ( + {getTranslatedLabel(item.data.attribute)}) + + + {t("information.pixels", { + ns: "common", + area: attributeAreaPx, + })}{" "} + · {attributeAreaPct}% + +
+ )} +
+ )} + + {item.data?.zones && item.data.zones.length > 0 && ( +
+ {item.data.zones.map((zone, zidx) => { + const color = getZoneColor(zone)?.join(",") ?? "0,0,0"; + return ( + { + e.stopPropagation(); + setSelectedZone(zone); + }} + style={{ + borderColor: `rgba(${color}, 0.6)`, + background: `rgba(${color}, 0.08)`, + }} + > + + + {item.data?.zones_friendly_names?.[zidx]} + + + ); + })} +
+ )}
diff --git a/web/src/types/frigateConfig.ts b/web/src/types/frigateConfig.ts index c7cbb50b8..985fe3457 100644 --- a/web/src/types/frigateConfig.ts +++ b/web/src/types/frigateConfig.ts @@ -305,6 +305,7 @@ export type CustomClassificationModelConfig = { enabled: boolean; name: string; threshold: number; + save_attempts?: number; object_config?: { objects: string[]; classification_type: string;