classifyre-cli 0.4.52__tar.gz → 0.4.53__tar.gz

{classifyre_cli-0.4.52 → classifyre_cli-0.4.53}/.turbo/turbo-build.log

@@ -1,3 +1,3 @@
 $ uv sync
-Resolved 293 packages in 218ms
+Resolved 292 packages in 384ms
 Checked 51 packages in 1ms

{classifyre_cli-0.4.52 → classifyre_cli-0.4.53}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: classifyre-cli
-Version: 0.4.52
+Version: 0.4.53
 Summary: Classifyre CLI — scan and classify unstructured data sources
 License: MIT
 Keywords: data,ingestion,metadata,pii,secrets,unstructured

{classifyre_cli-0.4.52 → classifyre_cli-0.4.53}/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@classifyre/cli",
-  "version": "0.4.52",
+  "version": "0.4.53",
   "private": true,
   "scripts": {
     "build": "uv sync",

{classifyre_cli-0.4.52 → classifyre_cli-0.4.53}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "classifyre-cli"
-version = "0.4.52"
+version = "0.4.53"
 description = "Classifyre CLI — scan and classify unstructured data sources"
 readme = "README.md"
 requires-python = ">=3.12"
@@ -128,6 +128,7 @@ detectors = [
     { include-group = "regex" },
     { include-group = "llm" },
     { include-group = "transcription" },
+    { include-group = "video" },
 ]
 file-processing = [
     "filetype>=1.2.0",
@@ -145,9 +146,16 @@ transcription = [
     # PyAV (no system ffmpeg required) and runs ctranslate2 for inference.
     "faster-whisper>=1.1.0",
 ]
+video = [
+    # Use the server-safe OpenCV 5 wheel. RapidOCR's desktop OpenCV dependency
+    # is excluded below because both distributions install the same cv2 module.
+    "opencv-python-headless>=5.0.0.93,<6.0.0",
+    "rapidocr-onnxruntime>=1.4.0",
+]
 ocr = [
     { include-group = "file-processing" },
     "docling>=2.94.0",
+    "opencv-python-headless>=5.0.0.93,<6.0.0",
     "rapidocr-onnxruntime>=1.4.0",
     # torchvision is a transitive dep of docling-ibm-models, but [tool.uv.sources]
     # index pins only apply to dependencies declared directly in this project.
@@ -407,6 +415,10 @@ markers = [
 
 [tool.uv]
 package = true
+# RapidOCR is the sole requester of opencv-python and works with the headless
+# distribution declared by our video and OCR groups. Installing both would
+# overwrite the same cv2 namespace and add GUI libraries to CPU-only containers.
+exclude-dependencies = ["opencv-python"]
 # Every platform we ship (desktop bundles + CI) must resolve to versions with
 # usable wheels. Without this, uv happily locks versions that are missing a
 # platform — e.g. deltalake 1.6.1 published no win_amd64 wheel, which broke the

{classifyre_cli-0.4.52 → classifyre_cli-0.4.53}/src/models/generated_input.py

@@ -112,14 +112,6 @@ class SamplingConfig(BaseModel):
         extra='forbid',
     )
     strategy: SamplingStrategy
-    enable_ocr: bool | None = Field(
-        False,
-        description='When true, enable OCR/text extraction for supported binary documents and images before routing text-capable detectors.',
-    )
-    enable_transcription: bool | None = Field(
-        False,
-        description='When true, transcribe audio and video files to text (via faster-whisper) before routing text-capable detectors. Slower and requires the transcription dependency.',
-    )
     order_by_column: str | None = Field(
         None,
         description='Column to use for LATEST sampling mode in tabular sources (usually created_at/updated_at). Auto-detected when not set.',
@@ -407,10 +399,6 @@ class YouTubeOptionalTranscript(BaseModel):
         None,
         description='Preferred caption language codes in priority order (e.g. ["en"]). Empty means accept any available language.',
     )
-    skip_transcript: bool | None = Field(
-        False,
-        description='When true, skip transcript fetching entirely (metadata-only assets, no detector content).',
-    )
 
 
 class YouTubeOptionalConnection(BaseModel):

{classifyre_cli-0.4.52 → classifyre_cli-0.4.53}/src/pipeline/detector_pipeline.py

@@ -108,11 +108,7 @@ class DetectorPipeline:
             return asset
 
         scan_started = datetime.now(UTC)
-        ocr_enabled = self.source.ocr_enabled()
-        transcription_enabled = self.source.transcription_enabled()
-        text_content_type = self._text_content_type_for_asset(
-            asset.asset_type, ocr_enabled, transcription_enabled
-        )
+        text_content_type = self._text_content_type_for_asset(asset.asset_type)
         link_content = self._build_links_payload(asset.links)
 
         text_detectors = []
@@ -765,8 +761,6 @@ class DetectorPipeline:
     def _text_content_type_for_asset(
         self,
         asset_type: OutputAssetType,
-        ocr_enabled: bool,
-        transcription_enabled: bool = False,
     ) -> str | None:
         mapping = {
             OutputAssetType.TXT: "text/plain",
@@ -775,9 +769,12 @@ class DetectorPipeline:
         }
         if asset_type in mapping:
             return mapping[asset_type]
-        if ocr_enabled and asset_type in {OutputAssetType.IMAGE, OutputAssetType.BINARY}:
-            return "text/plain"
-        if transcription_enabled and asset_type in {OutputAssetType.AUDIO, OutputAssetType.VIDEO}:
+        if asset_type in {
+            OutputAssetType.IMAGE,
+            OutputAssetType.BINARY,
+            OutputAssetType.AUDIO,
+            OutputAssetType.VIDEO,
+        }:
             return "text/plain"
         return None
 

{classifyre_cli-0.4.52 → classifyre_cli-0.4.53}/src/sources/base.py

@@ -325,18 +325,6 @@ class BaseSource(ABC):
             return stored.strip()
         return fallback_url
 
-    def ocr_enabled(self) -> bool:
-        """Return whether sampling-level OCR is enabled for this source."""
-        config = getattr(self, "config", None)
-        sampling = getattr(config, "sampling", None) if config is not None else None
-        return bool(getattr(sampling, "enable_ocr", False))
-
-    def transcription_enabled(self) -> bool:
-        """Return whether sampling-level audio/video transcription is enabled."""
-        config = getattr(self, "config", None)
-        sampling = getattr(config, "sampling", None) if config is not None else None
-        return bool(getattr(sampling, "enable_transcription", False))
-
     def parse_asset_bytes(
         self,
         file_bytes: bytes,
@@ -350,8 +338,6 @@ class BaseSource(ABC):
             file_bytes,
             declared_mime_type=declared_mime_type,
             file_name=file_name,
-            enable_ocr=self.ocr_enabled(),
-            enable_transcription=self.transcription_enabled(),
         )
 
     def iter_asset_pages(
@@ -371,8 +357,6 @@ class BaseSource(ABC):
             batch_size,
             include_column_names,
             file_name=file_name,
-            enable_ocr=self.ocr_enabled(),
-            enable_transcription=self.transcription_enabled(),
         )
 
     async def fetch_content_bytes(self, asset_id: str) -> tuple[bytes, str] | None:

{classifyre_cli-0.4.52 → classifyre_cli-0.4.53}/src/sources/object_storage/base.py

@@ -393,16 +393,11 @@ class ObjectStorageSourceBase(BaseSource, ABC):
         )
         normalized_mime = mime_type.split(";", 1)[0].strip().lower()
 
-        # Non-extractable types (images, opaque binary) carry no text. Audio/video
-        # are extractable only when transcription is enabled — otherwise they are
-        # treated as opaque binary. Everything else defers extraction to
-        # fetch_content_pages() so detectors receive content in configurable-sized
-        # pages instead of one monolithic blob.
-        is_media = normalized_mime.startswith(("audio/", "video/"))
-        is_non_extractable = (
-            normalized_mime.startswith("image/")
-            or (is_media and not self.transcription_enabled())
-            or normalized_mime in ("application/octet-stream", "application/zip")
+        # Images and audio/video are extractable automatically. Opaque binaries
+        # remain available to binary detectors but have no text page stream.
+        is_non_extractable = normalized_mime in (
+            "application/octet-stream",
+            "application/zip",
         )
 
         return ContentSnapshot(

{classifyre_cli-0.4.52 → classifyre_cli-0.4.53}/src/sources/youtube/source.py

@@ -6,10 +6,9 @@ Pipeline:
         -> video ids
         -> youtube-transcript-api (fetch captions)
             -> transcript found -> asset + metadata, transcript is detector content
-            -> no transcript:
-                -> sampling.enable_transcription set -> download audio (yt-dlp) and
-                   transcribe with faster-whisper -> transcript is detector content
-                -> otherwise -> asset + metadata only (no detector content)
+            -> download a bounded-resolution muxed stream
+                -> no captions -> transcribe audio with faster-whisper
+                -> sparse changed-frame OCR with OpenCV + RapidOCR
 
 Both libraries scrape public data and need no API key. ``yt-dlp`` and
 ``youtube-transcript-api`` are optional dependencies (``[youtube]`` group) and
@@ -89,6 +88,9 @@ class YouTubeSource(BaseSource):
         # Transcript text keyed by hash / external_url / video_id so the detector
         # pipeline (which probes both external_url and hash) resolves content.
         self._transcripts: dict[str, str] = {}
+        self._video_id_by_key: dict[str, str] = {}
+        self._assets_by_video_id: dict[str, SingleAssetScanResults] = {}
+        self._media_attempted: set[str] = set()
         self._cookiefile: str | None = None
 
     # ------------------------------------------------------------------
@@ -289,9 +291,8 @@ class YouTubeSource(BaseSource):
 
         Handles the documented failure cases (captions disabled, no captions,
         age-restricted/private, rate limiting) by logging and returning None so
-        the asset is still emitted without detector content. When no captions
-        exist, ``_build_video_asset`` falls back to ``_transcribe_audio`` if the
-        source has ``sampling.enable_transcription`` set.
+        the asset is still emitted. When no captions exist, the downloaded video
+        audio is always transcribed by ``_process_video_media``.
         """
         yt_mod = require_module("youtube_transcript_api", "YouTube", uv_groups=["youtube"])
 
@@ -331,17 +332,16 @@ class YouTubeSource(BaseSource):
             logger.warning("Transcript fetch failed for video %s: %s", video_id, exc)
             return None
 
-    def _download_audio(self, video_id: str, dest_dir: Path) -> Path | None:
-        """Download the best audio-only stream for a video into ``dest_dir``.
-
-        Returns the path to the downloaded file, or None on failure. No yt-dlp
-        post-processing is used, so the raw audio container (.m4a/.webm/…) is
-        written directly — faster-whisper decodes it via bundled PyAV, so no
-        system ffmpeg is required.
-        """
+    def _download_video(self, video_id: str, dest_dir: Path) -> Path | None:
+        """Download one muxed stream, preferring at most 480p to bound CPU/network."""
         opts = self._base_ydl_opts()
         opts["skip_download"] = False
-        opts["format"] = "bestaudio/best"
+        # Requiring both codecs avoids yt-dlp's ffmpeg merge path. Visual OCR only
+        # needs readable text, while the audio track remains suitable for Whisper.
+        opts["format"] = (
+            "best[height<=480][acodec!=none][vcodec!=none]/"
+            "best[acodec!=none][vcodec!=none]"
+        )
         opts["noplaylist"] = True
         opts["outtmpl"] = str(dest_dir / "%(id)s.%(ext)s")
         url = _WATCH_URL.format(video_id=video_id)
@@ -349,45 +349,53 @@ class YouTubeSource(BaseSource):
             with self._ydl_class()(opts) as ydl:
                 ydl.extract_info(url, download=True)
         except Exception as exc:
-            logger.warning("Failed to download audio for video %s: %s", video_id, exc)
+            logger.warning("Failed to download video %s: %s", video_id, exc)
             return None
         files = [p for p in dest_dir.iterdir() if p.is_file()]
         if not files:
-            logger.warning("Audio download produced no file for video %s", video_id)
+            logger.warning("Video download produced no file for video %s", video_id)
             return None
-        # bestaudio yields a single file; pick the largest if a sidecar slipped in.
+        # The selected format yields one file; pick the largest if a sidecar slipped in.
         return max(files, key=lambda p: p.stat().st_size)
 
-    def _transcribe_audio(self, video_id: str) -> _TranscriptResult | None:
-        """Download a video's audio and transcribe it with faster-whisper.
-
-        Used as a fallback when captions are unavailable. Returns None when the
-        download or transcription fails so the asset is still emitted.
-        """
-        from ...utils.transcription import transcribe_media
-
-        with tempfile.TemporaryDirectory(prefix="yt_audio_") as tmp:
-            path = self._download_audio(video_id, Path(tmp))
+    def _process_video_media(
+        self,
+        video_id: str,
+        *,
+        transcribe: bool,
+    ) -> tuple[_TranscriptResult | None, str]:
+        """Download once, then transcribe when needed and always OCR changed frames."""
+        from ...utils.transcription import transcribe_media_path
+        from ...utils.video_processing import extract_video_ocr_path
+
+        with tempfile.TemporaryDirectory(prefix="yt_video_") as tmp:
+            path = self._download_video(video_id, Path(tmp))
             if path is None:
-                return None
-            text, error = transcribe_media(
-                path.read_bytes(),
-                mime_type="",
-                file_name=path.name,
-            )
-        if error:
-            logger.warning("Whisper transcription failed for video %s: %s", video_id, error)
-            return None
-        if not text:
-            return None
-        logger.info("Transcribed video %s via faster-whisper (%d chars)", video_id, len(text))
-        return _TranscriptResult(
-            text=text,
-            language=None,
-            is_generated=True,
-            available_languages=[],
-            source="whisper",
-        )
+                return None, ""
+            visual_text, visual_error = extract_video_ocr_path(path)
+            if visual_error:
+                logger.warning("Visual OCR failed for YouTube video %s: %s", video_id, visual_error)
+
+            transcript: _TranscriptResult | None = None
+            if transcribe:
+                text, error = transcribe_media_path(
+                    path,
+                    mime_type="video/mp4",
+                )
+                if error:
+                    logger.warning("Whisper transcription failed for video %s: %s", video_id, error)
+                elif text:
+                    logger.info(
+                        "Transcribed video %s via faster-whisper (%d chars)", video_id, len(text)
+                    )
+                    transcript = _TranscriptResult(
+                        text=text,
+                        language=None,
+                        is_generated=True,
+                        available_languages=[],
+                        source="whisper",
+                    )
+            return transcript, visual_text
 
     # ------------------------------------------------------------------
     # Asset construction
@@ -417,18 +425,6 @@ class YouTubeSource(BaseSource):
         else:
             created_at = datetime.now(UTC)
 
-        transcript: _TranscriptResult | None = None
-        if not self._transcript_options().skip_transcript:
-            transcript = self._fetch_transcript(video_id)
-            # No captions: fall back to downloading the audio and transcribing it
-            # with faster-whisper when the source has transcription enabled.
-            if transcript is None and self.transcription_enabled():
-                transcript = self._transcribe_audio(video_id)
-        if transcript is not None:
-            self._transcripts[asset_hash] = transcript.text
-            self._transcripts[external_url] = transcript.text
-            self._transcripts[video_id] = transcript.text
-
         asset_metadata: dict[str, Any] = {
             "video_id": video_id,
             "title": title,
@@ -451,25 +447,17 @@ class YouTubeSource(BaseSource):
         if isinstance(upload_date, str) and upload_date:
             asset_metadata["upload_date"] = upload_date
 
-        asset_metadata["transcript_available"] = transcript is not None
-        if transcript is not None:
-            asset_metadata["transcript_source"] = transcript.source
-            if transcript.language:
-                asset_metadata["transcript_language"] = transcript.language
-            if transcript.is_generated is not None:
-                asset_metadata["transcript_is_generated"] = transcript.is_generated
-            if transcript.available_languages:
-                asset_metadata["caption_tracks"] = transcript.available_languages
+        asset_metadata["transcript_available"] = False
 
         # Checksum must only reflect content changes: view/like counters move on
         # every re-scan and would flip UNCHANGED assets to UPDATED spuriously.
         checksum_data = {
             "video_id": video_id,
             "title": title,
-            "transcript_available": asset_metadata["transcript_available"],
+            "detector_content": "",
         }
 
-        return SingleAssetScanResults(
+        asset = SingleAssetScanResults(
             hash=asset_hash,
             checksum=self.calculate_checksum(checksum_data),
             name=title,
@@ -482,6 +470,58 @@ class YouTubeSource(BaseSource):
             runner_id=self.runner_id,
             **self.metadata_fields("video", asset_metadata),
         )
+        self._assets_by_video_id[video_id] = asset
+        for key in (asset_hash, external_url, video_id):
+            self._video_id_by_key[key] = video_id
+
+        # Direct source usage (tests/local integrations) preserves eager behavior.
+        # The CLI's two-phase runner sets discovery_only and processes lazily in
+        # fetch_content(), bounded by max_concurrent_assets.
+        if not self._discovery_only:
+            self._populate_video_content(video_id)
+        return asset
+
+    def _populate_video_content(self, video_id: str) -> None:
+        if video_id in self._media_attempted:
+            return
+        self._media_attempted.add(video_id)
+
+        asset = self._assets_by_video_id.get(video_id)
+        if asset is None:
+            return
+        transcript = self._fetch_transcript(video_id)
+        media_transcript, visual_text = self._process_video_media(
+            video_id,
+            transcribe=transcript is None,
+        )
+        transcript = transcript or media_transcript
+        content_parts: list[str] = []
+        if transcript is not None:
+            content_parts.append(f"[Transcript]\n{transcript.text}")
+        if visual_text:
+            content_parts.append(visual_text)
+        detector_content = "\n\n".join(content_parts)
+
+        asset.metadata["transcript_available"] = transcript is not None
+        if transcript is not None:
+            asset.metadata["transcript_source"] = transcript.source
+            if transcript.language:
+                asset.metadata["transcript_language"] = transcript.language
+            if transcript.is_generated is not None:
+                asset.metadata["transcript_is_generated"] = transcript.is_generated
+            if transcript.available_languages:
+                asset.metadata["caption_tracks"] = transcript.available_languages
+
+        asset.checksum = self.calculate_checksum(
+            {
+                "video_id": video_id,
+                "title": asset.name,
+                "detector_content": detector_content,
+            }
+        )
+        if detector_content:
+            for key in (asset.hash, asset.external_url, video_id):
+                self._transcripts[key] = detector_content
 
     # ------------------------------------------------------------------
     # BaseSource interface
@@ -527,6 +567,9 @@ class YouTubeSource(BaseSource):
             return
 
         self._transcripts = {}
+        self._video_id_by_key = {}
+        self._assets_by_video_id = {}
+        self._media_attempted = set()
         video_ids = self._resolve_target_video_ids()
         logger.info("Resolved %d YouTube video(s) to extract", len(video_ids))
 
@@ -557,18 +600,28 @@ class YouTubeSource(BaseSource):
         return hash_id(type_value, asset_id)
 
     async def fetch_content(self, asset_id: str) -> tuple[str, str] | None:
-        """Return the cached transcript text for detector scanning.
+        """Lazily process and return transcript plus on-screen text."""
+        video_id = self._video_id_by_key.get(asset_id)
+        if video_id is not None:
+            import asyncio
 
-        Returns ``None`` when no transcript was found, so videos without captions
-        produce an asset + metadata but skip Phase 2 (detectors).
-        """
+            await asyncio.to_thread(self._populate_video_content, video_id)
         text = self._transcripts.get(asset_id)
         if not text:
             return None
         return text, text
 
     def evict_asset_cache(self, asset_hash: str) -> None:
+        video_id = self._video_id_by_key.pop(asset_hash, None)
         self._transcripts.pop(asset_hash, None)
+        if video_id:
+            asset = self._assets_by_video_id.pop(video_id, None)
+            if asset:
+                self._transcripts.pop(asset.external_url, None)
+                self._video_id_by_key.pop(asset.external_url, None)
+            self._transcripts.pop(video_id, None)
+            self._video_id_by_key.pop(video_id, None)
+            self._media_attempted.discard(video_id)
 
     def enrich_finding_location(
         self,
@@ -590,3 +643,6 @@ class YouTubeSource(BaseSource):
                 pass
             self._cookiefile = None
         self._transcripts.clear()
+        self._video_id_by_key.clear()
+        self._assets_by_video_id.clear()
+        self._media_attempted.clear()

{classifyre_cli-0.4.52 → classifyre_cli-0.4.53}/src/utils/dependency_groups.py

@@ -67,11 +67,4 @@ def recipe_uv_groups(recipe: dict[str, Any]) -> set[str]:
         detector_type = str(detector.get("type", "")).upper()
         groups |= DETECTOR_TYPE_GROUPS.get(detector_type, set())
 
-    sampling = recipe.get("sampling")
-    if isinstance(sampling, dict):
-        if sampling.get("enable_ocr"):
-            groups.add("ocr")
-        if sampling.get("enable_transcription"):
-            groups.add("transcription")
-
     return {group for group in groups if group}