API Reference¶

Comprehensive API documentation organized by task.

Core APIs¶

Detection API ¶

Face detection interfaces and utilities. - PrimateFaceProcessor: Main detection pipeline - MMDetection integration - YOLO detection - Batch processing utilities

Pose Estimation API ¶

Facial landmark detection and pose estimation. - PoseEstimator: Landmark detection - MMPose integration - Keypoint operations - Heatmap processing

Annotation API ¶

Interactive annotation and pseudo-labeling. - PseudoLabelGUI: Main annotation interface - SAM integration - Data converters - Quality control

Feature APIs¶

DINOv2 API ¶

Feature extraction and visualization. - DINOv2Extractor: Extract visual features - DiverseImageSelector: Smart subset selection - UMAP/t-SNE visualization - Attention maps

Converter API ¶

Cross-dataset landmark conversion. - LandmarkConverter: Format conversion - Neural network models - Training pipelines - Format mappings

Evaluation API ¶

Model comparison and metrics. - Performance metrics (NME, PCK, mAP) - Cross-framework comparison - Statistical analysis - Visualization tools

Module APIs¶

Demos Module ¶

Complete detection and pose pipeline. - End-to-end processing - Video support - Model management

GUI Module ¶

Interactive tools and utilities. - Annotation interfaces - Refinement tools - Batch operations

Quick Import Reference¶

# Detection and pose
from demos.process import PrimateFaceProcessor, PoseEstimator
from demos.utils import crop_face, expand_bbox

# Feature extraction
from dinov2.core import DINOv2Extractor
from dinov2.selection import DiverseImageSelector
from dinov2.visualization import plot_umap

# Landmark conversion  
from landmark_converter.apply_model import LandmarkConverter
from landmark_converter.train import train_converter

# Annotation
from gui.pseudolabel_gui_fm import PseudoLabelGUI
from gui.refine_boxes import COCORefinementGUI
from gui.core.sam import SAMAnnotator

# Evaluation
from evals.compare_pose_models import compare_models
from evals.core.metrics import calculate_nme, calculate_pck
from evals.visualize_eval_results import plot_comparison

API Patterns¶

Common Parameters¶

Most APIs follow these conventions: - device: "cuda:0" or "cpu" - batch_size: Processing batch size - num_workers: Parallel workers - verbose: Print progress

Error Handling¶

All APIs use consistent exception types:

try:
    result = api_call()
except ModelNotFoundError:
    # Handle missing model
except InvalidInputError:
    # Handle bad input
except ProcessingError:
    # Handle processing failure