IndoorLoc | 室内定位工具库
Multi-dataset, multi-model indoor localization toolkit

Docs · Installation · Quickstart · Datasets · Models · Contributing · Citation

English | 中文

---

Highlights

• Unified API for indoor localization across WiFi / BLE / CSI / UWB
• 12 verified datasets (auto-download when available) + extensible dataset registry
• Classic ML (scikit-learn) + deep models (PyTorch, timm)
• OpenMMLab-style YAML configs for reproducible experiments

Installation

GPU (CUDA 11.8)

conda create -n indoorloc python=3.10 pytorch torchvision pytorch-cuda=11.8 -c pytorch -c nvidia -y
conda activate indoorloc
pip install "indoorloc[full]"

CPU-only

conda create -n indoorloc python=3.10 pytorch torchvision cpuonly -c pytorch -y
conda activate indoorloc
pip install "indoorloc[full]"

Verify (optional)

python -c "import indoorloc, torch; print('indoorloc', indoorloc.__version__, '| torch', torch.__version__, '| cuda', torch.cuda.is_available())"

More install options (legacy, optional extras, troubleshooting): docs/installation.md.

---

Quickstart

Python API (recommended)

import indoorloc as iloc

train, test = iloc.load_dataset("ujindoorloc")            # 12 verified datasets
model = iloc.create_model("resnet18", dataset=train)      # Auto-configure model
results = model.fit(train).evaluate(test)                # Train & evaluate

Auto-download datasets · Auto-adapt dimensions · Auto-configure model

YAML Config + CLI

Config templates live in indoorloc/configs/.

indoorloc-train indoorloc/configs/wifi/resnet18_ujindoorloc.yaml

# Override any parameter
indoorloc-train indoorloc/configs/wifi/resnet18_ujindoorloc.yaml \
  --model.backbone.model_name efficientnet_b0 \
  --train.lr 5e-4 --train.epochs 200

# indoorloc/configs/wifi/resnet18_ujindoorloc.yaml
_base_:
  - ../_base_/models/resnet.yaml

model:
  backbone:
    model_name: resnet18
    pretrained: true
  head:
    num_floors: 5
    num_buildings: 3

train:
  epochs: 100
  lr: 0.001

Documentation

• Dataset catalogue (web): https://qdtiger.github.io/indoorloc/datasets.html
• Algorithm zoo (web): https://qdtiger.github.io/indoorloc/algorithms.html
• Config reference: indoorloc/configs/README.md

Datasets

Full dataset catalogue (web): https://qdtiger.github.io/indoorloc/datasets.html

• List available dataset IDs: iloc.list_available_datasets()
• Load a dataset: train, test = iloc.load_dataset("ujindoorloc")

Verified dataset IDs (12):

• WiFi: ujindoorloc, sodindoorloc, longtermwifi, tampere, wlanrssi, tuji1
• BLE: ble_indoor, ibeacon_rssi, ble_rssi_uci
• CSI: csi_fingerprint, hwild, haloc

Verified datasets (table)

Type	Dataset	ID	Samples
WiFi	UJIndoorLoc	ujindoorloc	21k
	SODIndoorLoc	sodindoorloc	24k
	LongTermWiFi	longtermwifi	104k
	Tampere	tampere	4.6k
	WLANRSSI	wlanrssi	2k
	TUJI1	tuji1	8.9k
BLE	BLEIndoor	ble_indoor	44k
	iBeaconRSSI	ibeacon_rssi	4.7k
	BLE RSSI UCI	ble_rssi_uci	1.4k
CSI	CSI Fingerprint	csi_fingerprint	489
	HWILD	hwild	409k
	HALOC	haloc	111k

Pending datasets (help wanted)

These datasets have download sources but are not yet integrated:

Dataset	Source	Notes
OpenCSI	Figshare	~2GB, format unverified
CSUIndoorLoc	GitHub	Format unverified
DICHASUS	DaRUS	14 scenarios
ESPARGOS	espargos.net	17-86GB
DeepMIMO	deepmimo.net	Requires pip install DeepMIMO
CSI2Pos	TIB	Requires login
CSI2TAoA	TIB	Requires login
MaMIMO CSI	IEEE DataPort	Requires account
WILDv2	Kaggle	Requires Kaggle API

Contribute: help us verify pending datasets! See CONTRIBUTING.md.

Models & Algorithms

Full algorithm zoo (web): https://qdtiger.github.io/indoorloc/algorithms.html

• List available models: iloc.list_models()
• Create a model: iloc.create_model("KNNLocalizer", k=5) / iloc.create_model("resnet18", dataset=train)

Algorithm families

Backed by sklearn (30+), timm (700+), lightly (10+), learn2learn (7+), and SKADA (20+).

• Supervised
  • Traditional ML: k-NN, WKNN, SVM, RF...
  • Deep: MLP, CNN1D, ResNet, ViT...
• Self-supervised
  • Contrastive: SimCLR, MoCo, NNCLR
  • Non-contrastive: BYOL, SimSiam, VICReg
• Meta-learning
  • Gradient-based: MAML, FOMAML, Reptile
  • Metric-based: ProtoNet, MatchingNet
• Transfer
  • Feature: CORAL, TCA
  • Reweight: KMM, KLIEP
  • Deep: DANN, MDD

Advanced usage

Custom model registration

import indoorloc as iloc
from indoorloc.registry import LOCALIZERS
from indoorloc.localizers.base import BaseLocalizer

@LOCALIZERS.register_module()
class MyLocalizer(BaseLocalizer):
    def fit(self, signals, locations, **kwargs):
        self._is_trained = True
        return self

    def predict(self, signal):
        raise NotImplementedError

model = iloc.create_model("MyLocalizer")

Project structure

indoorloc/
├── signals/          # WiFi, BLE, IMU, etc.
├── locations/        # Location classes
├── datasets/         # Verified + pending
├── localizers/       # ML & DL algorithms
├── evaluation/       # Metrics
└── configs/          # YAML configs

Evaluation metrics

Metric	Description
Mean Position Error	Average error (m)
Median Position Error	Median error (m)
Floor Accuracy	Floor classification
Building Accuracy	Building classification

Contributing

See CONTRIBUTING.md.

License

Apache License 2.0

Citation

@software{indoorloc,
  title = {IndoorLoc: A Unified Framework for Indoor Localization},
  year = {2024},
  url = {https://github.com/qdtiger/indoorloc}
}

Acknowledgements

• OpenMMLab — Registry and config system
• timm — 700+ pretrained models