pixelflow 0.1.2__tar.gz → 0.3.0__tar.gz

{pixelflow-0.1.2 → pixelflow-0.3.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pixelflow
-Version: 0.1.2
+Version: 0.3.0
 Summary: Computer vision processing library for visualizing and analyzing detection results from ML models
 Author-email: Emrah Nazif <emrah@datamarkin.com>, Datamarkin <support@datamarkin.com>
 License: MIT
@@ -8,7 +8,7 @@ Project-URL: Homepage, https://github.com/datamarkin/pixelflow
 Project-URL: Repository, https://github.com/datamarkin/pixelflow
 Project-URL: Issues, https://github.com/datamarkin/pixelflow/issues
 Project-URL: Documentation, https://pixelflow.datamarkin.com
-Keywords: computer-vision,visualization,annotation,post-processing,tracking,opencv,yolo,detectron2,ml-results
+Keywords: computer-vision,visualization,annotation,post-processing,tracking,opencv,yolo,detectron2,mayaku,ml-results
 Classifier: Development Status :: 3 - Alpha
 Classifier: Intended Audience :: Developers
 Classifier: Intended Audience :: Science/Research
@@ -29,6 +29,7 @@ Requires-Dist: shapely>=2.0.0
 Requires-Dist: numpy>=1.21.0
 Requires-Dist: opencv-python>=4.8.0
 Requires-Dist: scipy>=1.10.0
+Requires-Dist: Pillow>=9.0.0
 Provides-Extra: dev
 Requires-Dist: pytest>=7.0.0; extra == "dev"
 Requires-Dist: pytest-cov>=4.0.0; extra == "dev"
@@ -50,13 +51,13 @@ import pixelflow as pf
 from ultralytics import YOLO
 
 model = YOLO("yolo11n.pt")
-media = pf.Media("traffic.mp4")
+video = pf.VideoReader("traffic.mp4", width=640)
 tracker = pf.tracker.ByteTracker()
 zones = pf.Zones()
 zones.add_zone([(100, 400), (500, 400), (500, 600), (100, 600)], zone_id="entrance")
 
-for frame in media.frames:
-    detections = pf.detections.from_ultralytics(model.predict(frame))
+for frame in video:
+    detections = pf.from_ultralytics(model.predict(frame))
     detections = tracker.update(detections)
     zones.update(detections)
 
@@ -64,7 +65,8 @@ for frame in media.frames:
     frame = pf.annotate.label(frame, detections)
     frame = pf.annotate.zones(frame, zones)
 
-    pf.show_frame("Live", frame)
+    if pf.show_frame("Live", frame) == ord('q'):
+        break
 ```
 
 ## Installation
@@ -75,6 +77,12 @@ pip install pixelflow
 
 ## Core Concepts
 
+### Two Result Types
+
+Every model output becomes one of two things. **Detections** for anything that localises — detection, segmentation, keypoints, OCR. **Classifications** for anything that only names.
+
+They are peers, not variants of each other: three detections are three objects, while three classifications are three competing answers about one image.
+
 ### Detections - One Format, Every Framework
 
 Convert outputs from any ML framework into a unified format:
@@ -85,28 +93,80 @@ import pixelflow as pf
 # Ultralytics YOLO
 from ultralytics import YOLO
 model = YOLO("yolo11n.pt")
-detections = pf.detections.from_ultralytics(model.predict(image))
+detections = pf.from_ultralytics(model.predict(image))
 
 # Detectron2
 from detectron2.engine import DefaultPredictor
 predictor = DefaultPredictor(cfg)
-detections = pf.detections.from_detectron2(predictor(image), class_names=["person", "car"])
+detections = pf.from_detectron2(predictor(image), labels=["person", "car"])
+
+# Mayaku (returns Instances directly; take labels from the checkpoint)
+from mayaku import from_pretrained
+predictor = from_pretrained("mayaku-n-det")
+detections = pf.from_mayaku(predictor("photo.jpg"), labels=predictor.class_names)
 
 # HuggingFace Transformers
 from transformers import pipeline
 detector = pipeline("object-detection")
-detections = pf.detections.from_transformers(detector(image))
+detections = pf.from_transformers(detector(image))
 
-# Florence-2
-detections = pf.detections.from_florence2(model_output)
+# Florence-2 - captions and OCR reads land in det.text, <OD> classes in det.class_name
+detections = pf.from_florence2(model_output, task_prompt="<OD>")
 
 # SAM (Segment Anything)
-detections = pf.detections.from_sam(masks, scores)
+detections = pf.from_sam(masks, scores)
+
+# RF-DETR
+detections = pf.from_rfdetr(model_output)
+
+# Supervision
+detections = pf.from_supervision(sv_detections)
 
-# OCR Engines
-detections = pf.detections.from_tesseract(tesseract_output)
-detections = pf.detections.from_paddleocr(paddleocr_output)
-detections = pf.detections.from_easyocr(easyocr_output)
+# EasyOCR - text lands in det.text, the quad in det.segments
+import easyocr
+reader = easyocr.Reader(['en'])
+detections = pf.from_easyocr(reader.readtext("sign.jpg"))
+
+# Plain arrays (numpy or torch) - no framework container to convert from
+detections = pf.from_arrays(boxes, scores, class_ids, labels=model.class_names)
+```
+
+### Classifications - When the Model Only Names
+
+A classifier localises nothing, so its result carries no geometry at all:
+
+```python
+import pixelflow as pf
+
+# Ultralytics -cls checkpoints
+from ultralytics import YOLO
+model = YOLO("yolo11n-cls.pt")
+result = pf.from_ultralytics_classification(model.predict("dog.jpg"))
+
+result.top1.class_name              # 'golden retriever'
+result.top_k(5)                     # -> Classifications, best first
+result.to_json()
+
+# Plain scores - CLIP zero-shot, a vendored ResNet, anything with a score vector
+prompts = ["a photo of a cat", "a photo of a dog"]
+result = pf.from_scores(similarities, labels=prompts)
+
+# Multi-label: however many labels genuinely match, not a fixed count
+matches = result.filter_by_confidence(0.3)
+
+# Draw it
+frame = pf.annotate.classification(frame, result, top_k=3)
+```
+
+Scores are reported exactly as the model emitted them. PixelFlow does not normalise,
+re-softmax, or assume they sum to 1, so a softmax over a fixed class list and an
+independent per-label similarity both survive intact. A model emitting logits should
+be converted by the caller, who is the only one who knows which convention applies.
+
+Per-frame paths can skip building a row per class:
+
+```python
+result = pf.from_ultralytics_classification(model.predict(frame), top_k=5)
 ```
 
 ### Powerful Filtering
@@ -139,29 +199,32 @@ english_text = detections.filter_by_text_language("en")
 
 ### Media Handling
 
-Unified interface for videos, images, webcams, and streams:
+Purpose-built classes for videos, cameras, and images:
 
 ```python
 import pixelflow as pf
 
-# Video file with automatic resizing
-media = pf.Media("video.mp4", width=640)
-print(media.info)  # MediaInfo(resolution=1920x1080, fps=30.00, frames=900, duration=30.00s)
+# Read a video file
+video = pf.VideoReader("video.mp4", width=640)
+print(f"{video.width}x{video.height}, {video.fps}fps, {len(video)} frames")
 
-for frame in media.frames:  # Lazy loading, memory efficient
-    # Process frame
-    pf.show_frame("Preview", frame)
+for frame in video:       # Replayable — resets on each iteration
+    if pf.show_frame("Preview", frame) == ord('q'):
+        break
 
-# Webcam
-webcam = pf.Media(0)
+# Load a single image
+image = pf.read_image("photo.jpg", width=640)
 
-# Network stream
-stream = pf.Media("rtsp://camera.local/stream")
+# Webcam / network stream
+cam = pf.CameraStream(0, width=640)
+cam = pf.CameraStream("rtsp://camera.local/stream")
 
 # Write processed video
-for frame in pf.Media("input.mp4").frames:
+writer = pf.VideoWriter("output.mp4", fps=video.fps)
+for frame in video:
     processed = process(frame)
-    pf.write_frame("output.mp4", processed, media.info)
+    writer.write(processed)
+writer.close()
 ```
 
 ### Zone-Based Analytics
@@ -217,7 +280,7 @@ import pixelflow as pf
 tracker = pf.tracker.ByteTracker()
 
 for frame in media.frames:
-    detections = pf.detections.from_ultralytics(model.predict(frame))
+    detections = pf.from_ultralytics(model.predict(frame))
     detections = tracker.update(detections)
 
     for det in detections:
@@ -297,7 +360,7 @@ slicer = pf.SlicedInference(
 )
 
 def detector(image):
-    return pf.detections.from_ultralytics(model.predict(image))
+    return pf.from_ultralytics(model.predict(image))
 
 # Run on 4K image - automatically slices, detects, and merges
 large_image = cv2.imread("satellite.jpg")  # 4000x3000
@@ -320,7 +383,7 @@ crossings.add_line(
 )
 
 for frame in media.frames:
-    detections = tracker.update(pf.detections.from_ultralytics(model.predict(frame)))
+    detections = tracker.update(pf.from_ultralytics(model.predict(frame)))
     crossings.update(detections)
 
     print(f"Crossed: {crossings.get_counts()}")
@@ -347,15 +410,19 @@ report = detections.to_json_with_metrics()
 
 | Framework | Converter | Features |
 |-----------|-----------|----------|
+| None (plain numpy/torch arrays) | `from_arrays()` | Boxes, masks, keypoints |
 | Ultralytics (YOLO) | `from_ultralytics()` | Boxes, masks, keypoints |
 | Detectron2 | `from_detectron2()` | Boxes, masks, keypoints |
+| Mayaku | `from_mayaku()` | Boxes, masks, keypoints |
 | HuggingFace Transformers | `from_transformers()` | Boxes, scores |
-| Florence-2 | `from_florence2()` | Boxes, phrases |
+| Florence-2 | `from_florence2()` | Boxes, quads, polygons, captions |
 | SAM | `from_sam()` | Masks, scores |
-| Tesseract OCR | `from_tesseract()` | Text, boxes, confidence |
-| PaddleOCR | `from_paddleocr()` | Text, boxes, structure |
-| EasyOCR | `from_easyocr()` | Text, boxes |
-| PP-Structure | `from_ppstructure()` | Tables, formulas, layouts |
+| EfficientTAM | `from_efficienttam()` | Masks, scores |
+| RF-DETR | `from_rfdetr()` | Boxes, masks |
+| Supervision | `from_supervision()` | Boxes, masks, keypoints |
+| Falcon Perception | `from_falcon_perception()` | Boxes, masks |
+| EasyOCR | `from_easyocr()` | Quads, text, scores |
+| Datamarkin API | `from_datamarkin()` | Boxes, masks, keypoints |
 
 ## Documentation
 

{pixelflow-0.1.2 → pixelflow-0.3.0}/README.md

@@ -13,13 +13,13 @@ import pixelflow as pf
 from ultralytics import YOLO
 
 model = YOLO("yolo11n.pt")
-media = pf.Media("traffic.mp4")
+video = pf.VideoReader("traffic.mp4", width=640)
 tracker = pf.tracker.ByteTracker()
 zones = pf.Zones()
 zones.add_zone([(100, 400), (500, 400), (500, 600), (100, 600)], zone_id="entrance")
 
-for frame in media.frames:
-    detections = pf.detections.from_ultralytics(model.predict(frame))
+for frame in video:
+    detections = pf.from_ultralytics(model.predict(frame))
     detections = tracker.update(detections)
     zones.update(detections)
 
@@ -27,7 +27,8 @@ for frame in media.frames:
     frame = pf.annotate.label(frame, detections)
     frame = pf.annotate.zones(frame, zones)
 
-    pf.show_frame("Live", frame)
+    if pf.show_frame("Live", frame) == ord('q'):
+        break
 ```
 
 ## Installation
@@ -38,6 +39,12 @@ pip install pixelflow
 
 ## Core Concepts
 
+### Two Result Types
+
+Every model output becomes one of two things. **Detections** for anything that localises — detection, segmentation, keypoints, OCR. **Classifications** for anything that only names.
+
+They are peers, not variants of each other: three detections are three objects, while three classifications are three competing answers about one image.
+
 ### Detections - One Format, Every Framework
 
 Convert outputs from any ML framework into a unified format:
@@ -48,28 +55,80 @@ import pixelflow as pf
 # Ultralytics YOLO
 from ultralytics import YOLO
 model = YOLO("yolo11n.pt")
-detections = pf.detections.from_ultralytics(model.predict(image))
+detections = pf.from_ultralytics(model.predict(image))
 
 # Detectron2
 from detectron2.engine import DefaultPredictor
 predictor = DefaultPredictor(cfg)
-detections = pf.detections.from_detectron2(predictor(image), class_names=["person", "car"])
+detections = pf.from_detectron2(predictor(image), labels=["person", "car"])
+
+# Mayaku (returns Instances directly; take labels from the checkpoint)
+from mayaku import from_pretrained
+predictor = from_pretrained("mayaku-n-det")
+detections = pf.from_mayaku(predictor("photo.jpg"), labels=predictor.class_names)
 
 # HuggingFace Transformers
 from transformers import pipeline
 detector = pipeline("object-detection")
-detections = pf.detections.from_transformers(detector(image))
+detections = pf.from_transformers(detector(image))
 
-# Florence-2
-detections = pf.detections.from_florence2(model_output)
+# Florence-2 - captions and OCR reads land in det.text, <OD> classes in det.class_name
+detections = pf.from_florence2(model_output, task_prompt="<OD>")
 
 # SAM (Segment Anything)
-detections = pf.detections.from_sam(masks, scores)
+detections = pf.from_sam(masks, scores)
+
+# RF-DETR
+detections = pf.from_rfdetr(model_output)
+
+# Supervision
+detections = pf.from_supervision(sv_detections)
 
-# OCR Engines
-detections = pf.detections.from_tesseract(tesseract_output)
-detections = pf.detections.from_paddleocr(paddleocr_output)
-detections = pf.detections.from_easyocr(easyocr_output)
+# EasyOCR - text lands in det.text, the quad in det.segments
+import easyocr
+reader = easyocr.Reader(['en'])
+detections = pf.from_easyocr(reader.readtext("sign.jpg"))
+
+# Plain arrays (numpy or torch) - no framework container to convert from
+detections = pf.from_arrays(boxes, scores, class_ids, labels=model.class_names)
+```
+
+### Classifications - When the Model Only Names
+
+A classifier localises nothing, so its result carries no geometry at all:
+
+```python
+import pixelflow as pf
+
+# Ultralytics -cls checkpoints
+from ultralytics import YOLO
+model = YOLO("yolo11n-cls.pt")
+result = pf.from_ultralytics_classification(model.predict("dog.jpg"))
+
+result.top1.class_name              # 'golden retriever'
+result.top_k(5)                     # -> Classifications, best first
+result.to_json()
+
+# Plain scores - CLIP zero-shot, a vendored ResNet, anything with a score vector
+prompts = ["a photo of a cat", "a photo of a dog"]
+result = pf.from_scores(similarities, labels=prompts)
+
+# Multi-label: however many labels genuinely match, not a fixed count
+matches = result.filter_by_confidence(0.3)
+
+# Draw it
+frame = pf.annotate.classification(frame, result, top_k=3)
+```
+
+Scores are reported exactly as the model emitted them. PixelFlow does not normalise,
+re-softmax, or assume they sum to 1, so a softmax over a fixed class list and an
+independent per-label similarity both survive intact. A model emitting logits should
+be converted by the caller, who is the only one who knows which convention applies.
+
+Per-frame paths can skip building a row per class:
+
+```python
+result = pf.from_ultralytics_classification(model.predict(frame), top_k=5)
 ```
 
 ### Powerful Filtering
@@ -102,29 +161,32 @@ english_text = detections.filter_by_text_language("en")
 
 ### Media Handling
 
-Unified interface for videos, images, webcams, and streams:
+Purpose-built classes for videos, cameras, and images:
 
 ```python
 import pixelflow as pf
 
-# Video file with automatic resizing
-media = pf.Media("video.mp4", width=640)
-print(media.info)  # MediaInfo(resolution=1920x1080, fps=30.00, frames=900, duration=30.00s)
+# Read a video file
+video = pf.VideoReader("video.mp4", width=640)
+print(f"{video.width}x{video.height}, {video.fps}fps, {len(video)} frames")
 
-for frame in media.frames:  # Lazy loading, memory efficient
-    # Process frame
-    pf.show_frame("Preview", frame)
+for frame in video:       # Replayable — resets on each iteration
+    if pf.show_frame("Preview", frame) == ord('q'):
+        break
 
-# Webcam
-webcam = pf.Media(0)
+# Load a single image
+image = pf.read_image("photo.jpg", width=640)
 
-# Network stream
-stream = pf.Media("rtsp://camera.local/stream")
+# Webcam / network stream
+cam = pf.CameraStream(0, width=640)
+cam = pf.CameraStream("rtsp://camera.local/stream")
 
 # Write processed video
-for frame in pf.Media("input.mp4").frames:
+writer = pf.VideoWriter("output.mp4", fps=video.fps)
+for frame in video:
     processed = process(frame)
-    pf.write_frame("output.mp4", processed, media.info)
+    writer.write(processed)
+writer.close()
 ```
 
 ### Zone-Based Analytics
@@ -180,7 +242,7 @@ import pixelflow as pf
 tracker = pf.tracker.ByteTracker()
 
 for frame in media.frames:
-    detections = pf.detections.from_ultralytics(model.predict(frame))
+    detections = pf.from_ultralytics(model.predict(frame))
     detections = tracker.update(detections)
 
     for det in detections:
@@ -260,7 +322,7 @@ slicer = pf.SlicedInference(
 )
 
 def detector(image):
-    return pf.detections.from_ultralytics(model.predict(image))
+    return pf.from_ultralytics(model.predict(image))
 
 # Run on 4K image - automatically slices, detects, and merges
 large_image = cv2.imread("satellite.jpg")  # 4000x3000
@@ -283,7 +345,7 @@ crossings.add_line(
 )
 
 for frame in media.frames:
-    detections = tracker.update(pf.detections.from_ultralytics(model.predict(frame)))
+    detections = tracker.update(pf.from_ultralytics(model.predict(frame)))
     crossings.update(detections)
 
     print(f"Crossed: {crossings.get_counts()}")
@@ -310,15 +372,19 @@ report = detections.to_json_with_metrics()
 
 | Framework | Converter | Features |
 |-----------|-----------|----------|
+| None (plain numpy/torch arrays) | `from_arrays()` | Boxes, masks, keypoints |
 | Ultralytics (YOLO) | `from_ultralytics()` | Boxes, masks, keypoints |
 | Detectron2 | `from_detectron2()` | Boxes, masks, keypoints |
+| Mayaku | `from_mayaku()` | Boxes, masks, keypoints |
 | HuggingFace Transformers | `from_transformers()` | Boxes, scores |
-| Florence-2 | `from_florence2()` | Boxes, phrases |
+| Florence-2 | `from_florence2()` | Boxes, quads, polygons, captions |
 | SAM | `from_sam()` | Masks, scores |
-| Tesseract OCR | `from_tesseract()` | Text, boxes, confidence |
-| PaddleOCR | `from_paddleocr()` | Text, boxes, structure |
-| EasyOCR | `from_easyocr()` | Text, boxes |
-| PP-Structure | `from_ppstructure()` | Tables, formulas, layouts |
+| EfficientTAM | `from_efficienttam()` | Masks, scores |
+| RF-DETR | `from_rfdetr()` | Boxes, masks |
+| Supervision | `from_supervision()` | Boxes, masks, keypoints |
+| Falcon Perception | `from_falcon_perception()` | Boxes, masks |
+| EasyOCR | `from_easyocr()` | Quads, text, scores |
+| Datamarkin API | `from_datamarkin()` | Boxes, masks, keypoints |
 
 ## Documentation
 

pixelflow-0.3.0/pixelflow/__init__.py

@@ -0,0 +1,91 @@
+__version__ = "0.3.0"
+__author__ = "Datamarkin"
+
+# Import core modules
+from . import media
+from . import annotators as annotate
+from . import transforms as transform
+from . import colors
+from . import slicer
+from . import smoother
+from . import timer
+from . import tracker
+from . import assets
+
+# Result types and their converters. Every model output PixelFlow understands becomes
+# one of two things: Detections for anything that localises, Classifications for
+# anything that only names. They are peers, not variants of each other.
+#
+# Converters are flat -- pf.from_ultralytics(...) -- and each subpackage owns its own
+# export list, so a new one is declared once rather than re-typed here. Detection is
+# the unmarked case because almost every model that produces a container localises
+# something; only classification needs the suffix.
+from .detections import *
+from .detections import __all__ as _DETECTION_EXPORTS
+from .classifications import *
+from .classifications import __all__ as _CLASSIFICATION_EXPORTS
+
+# Import specific functions for top-level access
+from .media import (
+    DisplayExit, VideoReader, CameraStream, VideoWriter,
+    read_image, read_video, display_video, display_image, save_image,
+    close_display, to_pil, from_pil,
+)
+from .zones import Zones
+from .crossings import Crossings
+from .slicer import SlicedInference, auto_slice_size
+from .buffer import Buffer
+from .smoother import smooth
+from .timer import TimeTracker
+
+# Define the public API
+__all__ = [
+    # Result types and converters, declared by the subpackages that own them
+    *_DETECTION_EXPORTS,
+    *_CLASSIFICATION_EXPORTS,
+
+    # Visual components
+    "annotate",
+    "colors",
+
+    # Image transformations
+    "transform",
+
+    # Media handling
+    "media",
+    "DisplayExit",
+    "VideoReader",
+    "CameraStream",
+    "VideoWriter",
+    "read_image",
+    "read_video",
+    "display_video",
+    "display_image",
+    "save_image",
+    "close_display",
+    "to_pil",
+    "from_pil",
+
+    # Spatial analysis
+    "zones",
+    "Zones",
+    "crossings",
+    "Crossings",
+
+    # Processing utilities
+    "assets",
+    "slicer",
+    "SlicedInference",
+    "auto_slice_size",
+    "Buffer",
+    "smooth",
+
+    # Performance & tracking
+    "timer",
+    "TimeTracker",
+    "tracker",
+
+    # Metadata
+    "__version__",
+    "__author__"
+]

{pixelflow-0.1.2 → pixelflow-0.3.0}/pixelflow/annotators/__init__.py

@@ -12,6 +12,7 @@ allowing easy chaining of annotation operations.
 from .anchors import anchors
 from .blur import blur
 from .box import box
+from .classification import classification
 from .label import label
 from .crossing import crossings
 from .mask import mask
@@ -23,7 +24,7 @@ from .zones import zones
 # TODO: Import these once implemented
 from .filled_box import filled_box
 # from .dot import dot
-# from .fps_counter import fps_counter
+from .fps_counter import fps_counter
 from .grid_overlay import grid_overlay
 # from .heatmap import heatmap
 from .keypoint import keypoint
@@ -41,6 +42,7 @@ __all__ = [
     # Core annotators
     'anchors',
     'box',
+    'classification',
     'label',
     'mask',
     'keypoint',
@@ -56,5 +58,8 @@ __all__ = [
 
     # Zone annotators
     'zones',
-    'crossings'
+    'crossings',
+
+    # Overlay annotators
+    'fps_counter',
 ]

{pixelflow-0.1.2 → pixelflow-0.3.0}/pixelflow/annotators/anchors.py

@@ -38,7 +38,7 @@ def anchors(
     Supports drawing individual anchor points or multiple points simultaneously.
     
     Args:
-        image (np.ndarray): Input image to draw anchor points on in BGR format.
+        image (np.ndarray): Input image to draw anchor points on in RGB format.
                            Image is modified in-place.
         detections (Detections): Detection results containing bounding boxes.
                                 Each detection must have a 'bbox' attribute with (x1, y1, x2, y2) coordinates.
@@ -51,7 +51,7 @@ def anchors(
         thickness (Optional[int]): Thickness of circle outline in pixels.
                                   Use -1 for filled circles.
                                   If None, defaults to -1 (filled).
-        colors (Optional[List[tuple]]): List of BGR color tuples for custom colors.
+        colors (Optional[List[tuple]]): List of RGB color tuples for custom colors.
                                        Colors mapped to unique class_ids in order of appearance.
                                        If None, uses default ColorManager colors.
     
@@ -71,7 +71,7 @@ def anchors(
         >>> image = cv2.imread("path/to/image.jpg")
         >>> model = YOLO("yolo11n.pt")
         >>> outputs = model.predict(image)
-        >>> detections = pf.results.from_ultralytics(outputs)
+        >>> detections = pf.from_ultralytics(outputs)
         >>> 
         >>> # Draw all main anchor points (default)
         >>> annotated = pf.annotators.anchors(image, detections)

{pixelflow-0.1.2 → pixelflow-0.3.0}/pixelflow/annotators/blur.py

@@ -61,7 +61,7 @@ def blur(
         >>> image = cv2.imread("faces_in_crowd.jpg")
         >>> model = YOLO("yolo11n.pt")
         >>> outputs = model.predict(image)  # Raw model outputs
-        >>> detections = pf.results.from_ultralytics(outputs)  # Convert to PixelFlow format
+        >>> detections = pf.from_ultralytics(outputs)  # Convert to PixelFlow format
         >>> 
         >>> # Apply blur with default adaptive settings
         >>> blurred_image = pf.annotators.blur(image, detections)

{pixelflow-0.1.2 → pixelflow-0.3.0}/pixelflow/annotators/box.py

@@ -22,13 +22,13 @@ def box(
     Automatically adapts line thickness based on image dimensions for optimal visibility.
 
     Args:
-        image (np.ndarray): Input image to draw boxes on (BGR format).
+        image (np.ndarray): Input image to draw boxes on (RGB format).
                            Modified in-place with drawn bounding boxes.
         detections (Detections): Detections object containing bounding boxes.
                                 Each detection must have a 'bbox' attribute with (x1, y1, x2, y2) coordinates.
         thickness (Optional[int]): Line thickness for bounding boxes in pixels.
                                   If None, automatically determined based on image size.
-        colors (Optional[List[tuple]]): List of BGR color tuples to override default colors.
+        colors (Optional[List[tuple]]): List of RGB color tuples to override default colors.
                                        Colors are mapped to unique class_ids in order of appearance.
                                        If None, uses default ColorManager colors.
 
@@ -48,13 +48,13 @@ def box(
         >>> image = cv2.imread("path/to/image.jpg")
         >>> model = YOLO("yolo11n.pt")
         >>> outputs = model.predict(image)  # Raw model outputs
-        >>> detections = pf.results.from_ultralytics(outputs)  # Convert to PixelFlow format
+        >>> detections = pf.from_ultralytics(outputs)  # Convert to PixelFlow format
         >>>
         >>> # Draw boxes with default colors and adaptive thickness
         >>> annotated = pf.annotators.box(image, detections)
         >>>
         >>> # Override with custom colors for specific classes
-        >>> custom_colors = [(255, 0, 0), (0, 255, 0), (0, 0, 255)]  # Blue, Green, Red
+        >>> custom_colors = [(255, 0, 0), (0, 255, 0), (0, 0, 255)]  # Red, Green, Blue
         >>> annotated = pf.annotators.box(image, detections, colors=custom_colors)
         >>>
         >>> # Use custom thickness for fine control