API Reference¶

Welcome to the visdet API reference documentation.

Overview¶

VisDet provides a comprehensive API for object detection tasks. The API is organized into several key modules:

Core Modules¶

Detection APIs ¶

High-level APIs for model initialization, training, and inference.

init_detector: Initialize a detector from config
inference_detector: Run inference on images
train_detector: Train a detection model
show_result_pyplot: Visualize detection results

Models ¶

Model architectures and components.

Detectors: Two-stage and single-stage detectors
Backbones: ResNet, ResNeXt, Swin Transformer, etc.
Necks: FPN, PAFPN, etc.
Heads: Detection heads for various architectures

Datasets ¶

Dataset classes and data pipelines.

Dataset Classes: COCO, VOC, custom datasets
Transforms: Data augmentation and preprocessing
Loaders: DataLoader utilities

Quick Examples¶

Initialize a Model¶

from mmdet.apis import init_detector

config_file = 'configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco.py'
checkpoint_file = 'checkpoints/faster_rcnn_r50_fpn_1x_coco.pth'

model = init_detector(config_file, checkpoint_file, device='cuda:0')

Run Inference¶

from mmdet.apis import inference_detector, show_result_pyplot

img = 'demo/demo.jpg'
result = inference_detector(model, img)
show_result_pyplot(model, img, result, score_thr=0.3)

Work with Datasets¶

from mmdet.datasets import build_dataset
from mmcv import Config

cfg = Config.fromfile('config.py')
dataset = build_dataset(cfg.data.train)

Module Organization¶

mmdet/
├── apis/          # High-level APIs
├── models/        # Model architectures
│   ├── detectors/ # Detector implementations
│   ├── backbones/ # Backbone networks
│   ├── necks/     # Neck modules
│   └── heads/     # Detection heads
├── datasets/      # Dataset classes
│   ├── pipelines/ # Data transforms
│   └── samplers/  # Data samplers
├── core/          # Core utilities
└── utils/         # Helper functions

Additional Resources¶

API Stability¶

The API follows semantic versioning. While we strive to maintain backward compatibility, some APIs may change between major versions. Deprecated functions will include warnings with migration guidance.