Skip to content

README.md

Latest commit

 

History

History
484 lines (371 loc) · 32.6 KB

README.md

File metadata and controls

484 lines (371 loc) · 32.6 KB

Tiny ML ModelZoo

Welcome to the Tiny ML ModelZoo - Texas Instruments' central repository for AI models, examples, and configurations for microcontroller (MCU) applications.

Table of Contents

Introduction

Texas Instruments MCU AI Toolchain

Texas Instruments provides a comprehensive toolchain for developing, training, and deploying machine learning models on resource-constrained microcontrollers. The toolchain consists of three main components:

Detailed User Guide: TI Tiny ML Tensorlab User Guide

Component Purpose Who Should Use It
tinyml-modelzoo (this repo) Models, examples, and configurations End customers - This is your starting point
tinyml-modelmaker Training orchestration and compilation Developers extending the toolchain
tinyml-tinyverse Core training scripts and utilities Advanced developers only
tinyml-modeloptimization Quantization scripts and utilities Advanced developers only

As an end customer, you only need to work with tinyml-modelzoo. The example configurations here will automatically use the underlying toolchain components to train and compile models for your target MCU.

What's in ModelZoo?

tinyml-modelzoo/
├── examples/              # Ready-to-run example configurations
├── tinyml_modelzoo/
│   ├── models/            # Neural network model definitions
│   ├── model_descriptions/ # Model metadata for GUI integration
│   └── device_info/       # Target device performance data
├── run_tinyml_modelzoo.sh        # Training wrapper (Linux)
├── run_tinyml_modelzoo.bat       # Training wrapper (Windows)
└── ADDING_NEW_MODELS.md   # Guide for adding custom models

Supported Target Devices

C2000 DSP Family (Texas Instruments)

Device NPU Description Notes
F28P55 Yes C2000 32-bit MCU Recommended for complex models
F28P65 No C2000 32-bit MCU, 150 MHz High performance
F29H85 No C2000 64-bit MCU with C29x core High capacity
F29P58 No C2000 64-bit MCU with C29x core High capacity
F29P32 No C2000 64-bit MCU with C29x core High capacity
F2837 No C2000 32-bit dual-core MCU, 200 MHz General purpose
F28003 No C2000 32-bit MCU, 100 MHz Cost-optimized
F28004 No C2000 32-bit MCU, 100 MHz Cost-optimized
F280013 No C2000 32-bit MCU, 100 MHz Entry-level
F280015 No C2000 32-bit MCU, 120 MHz Entry-level

MSPM0 Family (Arm Cortex-M0+)

Device NPU Description Notes
MSPM0G3507 No 80 MHz Arm Cortex-M0+ Ultra-low power, classification only
MSPM0G3519 No 80 MHz Arm Cortex-M0+ Ultra-low power
MSPM0G5187 Yes 80 MHz Arm Cortex-M0+ Ultra-low power, NPU-accelerated

MSPM33C Family (Arm Cortex-M33)

Device NPU Description Notes
MSPM33C32 No 160 MHz Arm Cortex-M33, TrustZone 1MB flash, 256kB SRAM
MSPM33C34 No 160 MHz Arm Cortex-M33 High performance

AM13 Family (Arm Cortex-M33)

Device NPU Description Notes
AM13E2 Yes Arm Cortex-M33 MCU NPU-accelerated

AM26x Family (Arm Cortex-R5)

Device NPU Description Notes
AM263 No Quad-core Arm Cortex-R5F, 400 MHz High performance
AM263P No Quad-core Arm Cortex-R5F, 400 MHz High performance
AM261 No Single-core Arm Cortex-R5F, 400 MHz Cost-optimized

Connectivity Devices (Wireless)

Device Hardware Accelerator Description Notes
CC2755 CDE 96 MHz Arm Cortex-M33 wireless MCU Optimized for PIR/wireless apps
CC1312 No Arm Cortex-M4F wireless MCU Sub-1GHz
CC1314 No Arm Cortex-M33 wireless MCU Sub-1GHz
CC1352 No Arm Cortex-M4F wireless MCU Sub-1GHz
CC1354 No Arm Cortex-M33 wireless MCU Sub-1GHz
CC35X1 CDE Arm Cortex-M33 wireless MCU Wi-Fi + BLE combo

Quick Start

Prerequisites

  1. Python 3.10 environment with the Tiny ML toolchain installed
  2. Clone the tinyml-tensorlab repository

Running an Example

Linux:

# Activate your Python environment

# Navigate to modelzoo
cd tinyml-modelzoo

# Run an example (e.g., generic_timeseries_classification)
./run_tinyml_modelzoo.sh examples/generic_timeseries_classification/config.yaml

Windows:

# Navigate to modelzoo
cd tinyml-modelzoo

# Run an example
run_tinyml_modelzoo.bat examples\generic_timeseries_classification\config.yaml

What Happens When You Run an Example?

  1. Dataset Download - The toolchain downloads the required dataset (if not already present)
  2. Data Processing - Feature extraction and preprocessing are applied
  3. Model Training - The neural network is trained on your data
  4. Quantization - The model is optimized for MCU deployment
  5. Compilation - TI's Neural Network Compiler generates device-ready code

  • Output artifacts are saved to ../tinyml-modelmaker/data/projects/<project_name>/.

  • You can choose to save the output artifacts in your own custom directory, by specifying in the respective config.yaml by adding this under the common section:

    common:
  projects_path: './your/choice'  # or absolute path
  # ... other settings

Supported Task Categories

Tiny ML ModelZoo supports the following AI task categories:

Task Category Description Use Cases
Time Series Classification Categorize time-series data into discrete classes Fault detection, activity recognition, anomaly classification
Time Series Regression Predict continuous values from time-series inputs Torque estimation, speed prediction, load measurement
Time Series Forecasting Predict future values based on historical patterns Temperature prediction, demand forecasting
Time Series Anomaly Detection Identify abnormal patterns using autoencoder-based models Equipment health monitoring, predictive maintenance
Image Classification Categorize images into classes Visual inspection, object recognition

Understanding Each Task

Classification - The model outputs a probability distribution over predefined classes. Best for: "Is this a A fault or B fault or C fault?", "Which type of activity is this?"

Regression - The model outputs a continuous numerical value. Best for: "What is the current torque?", "What will the temperature be?"

Forecasting - Predicts future values in a time series. Best for: "What will happen next?"

Anomaly Detection - Uses autoencoders to learn "normal" patterns. Reconstruction error indicates anomalies. Best for: "Is this behavior normal?"

  • The main difference between Anomaly Detection v/s Classification can be understood with the below example:

    • Is it Normal? or an anomaly? --> Anomaly Detection (binary outcome)
    • Is it Normal? or anomaly type A? or anomaly type B? or anomaly type C? --> Classification (multiple categories)

  • The main difference between Classification v/s Regression can be understood with the below example:

    • Using independent variables Xa, Xb, Xc to predict dependent discrete variable (target) Y --> Classification
      • Y can produce discrete values that indicate if it stands for Class A / Class B / Class C .... so on
    • Using independent variables Xa, Xb, Xc to predict dependent continuous variable (target) Y --> Regression

  • The main difference between Regression v/s Forecasting can be understood with the below example:

    • Using independent variables Xa, Xb, Xc to predict dependent continuous variable (target) Y at the same time instant --> Regression
    • Using independent variables Xa, Xb, Xc to predict dependent continuous variable (target) Xa (or Xb or Xc) for the next time instant--> Forecasting

There are two ways to proceed using this toolchain.

  1. If you, as a user, find that there is an application under the Example Applications section below, then you can proceed with it.
  2. However, if you do not find any applications that are of your direct interest, you may as well use the toolchain to do either of the Supported Task Categories as mentioned above referring to the generic example for each of them:
Generic Example Type Example Description
Timeseries Classification generic_timeseries_classification Classify sine/square/sawtooth waveforms. Start here to learn the toolchain.
Timeseries Regression generic_timeseries_regression Generic regression example for continuous value prediction.
Timeseries Forecasting generic_timeseries_forecasting Generic forecasting example for time series prediction.
Timeseries Anomaly Detection generic_timeseries_anomalydetection Generic anomaly detection example using autoencoders.

Example Applications

The following ready-to-use examples demonstrate various AI applications for MCUs, organized by task type.

Generic Timeseries Applications

These applications use generic task types that can be adapted to your custom datasets. All generic timeseries applications support all 22 target devices.

All 22 generic timeseries applications support all 22 devices: F280013, F280015, F28003, F28004, F2837, F28P55, F28P65, F29P58, F29P32, MSPM0G3507, MSPM0G3519, MSPM0G5187, MSPM33C32, F29H85, CC2755, CC1352, CC1354, CC35X1, AM263, AM263P, AM261, AM13E2

Example Name Task Type Description
generic_timeseries_classification generic_timeseries_classification Classify sine/square/sawtooth waveforms - Start here to learn the toolchain
generic_timeseries_regression generic_timeseries_regression Generic regression example for continuous value prediction
generic_timeseries_forecasting generic_timeseries_forecasting Generic forecasting example for time series prediction
generic_timeseries_anomalydetection generic_timeseries_anomalydetection Generic anomaly detection example using autoencoders
The below examples demonstrate the above AI task types with real world data:
branched_model_parameters generic_timeseries_classification Human Activity Recognition from accelerometer/gyroscope data
electrical_fault generic_timeseries_classification Classify transmission line faults using voltage and current (2-class and 6-class variants)
gas_sensor generic_timeseries_classification Identify gas type and concentration from sensor array data
grid_fault_detection generic_timeseries_classification Detect electrical grid faults from sensor data
grid_stability generic_timeseries_classification Predict power grid stability from node parameters
nilm_appliance_usage_classification generic_timeseries_classification Non-Intrusive Load Monitoring - identify active appliances
PLAID_nilm_classification generic_timeseries_classification Appliance identification using the PLAID dataset
induction_motor_speed_prediction generic_timeseries_regression Predict induction motor speed from electrical signals
mosfet_temp_prediction generic_timeseries_regression Predict MOSFET temperature from electrical parameters
reg_washing_machine generic_timeseries_regression Predict washing machine load weight
torque_measurement_regression generic_timeseries_regression Predict PMSM motor torque from current measurements
forecasting_pmsm_rotor_temp generic_timeseries_forecasting Forecast PMSM rotor winding temperature
hvac_indoor_temp_forecast generic_timeseries_forecasting Predict indoor temperature for HVAC control
dc_arc_fault (anomaly detection - DSI) generic_timeseries_anomalydetection Detect anomalous DC arc patterns using autoencoder (DSI dataset)
dc_arc_fault (anomaly detection - DSK) generic_timeseries_anomalydetection Detect anomalous DC arc patterns using autoencoder (DSK dataset)
ecg_classification (anomaly detection) generic_timeseries_anomalydetection Detect anomalous heartbeat patterns from ECG signals
fan_blade_fault_classification (anomaly detection) generic_timeseries_anomalydetection Detect anomalous fan blade behavior from accelerometer data
motor_bearing_fault (anomaly detection) generic_timeseries_anomalydetection Detect anomalous bearing behavior from vibration data

Application-Specific Examples

These applications are designed for specific use cases with optimized models and datasets.

Example Name Task Type Supported Devices Description
ac_arc_fault arc_fault F280013, F280015, F28003, F28004, F2837, F28P55, F28P65, MSPM0G3507, MSPM0G3519, MSPM0G5187, MSPM33C32, F29H85, AM13E2, AM263 Detect AC arc faults in electrical systems
dc_arc_fault arc_fault F280013, F280015, F28003, F28004, F2837, F28P55, F28P65, MSPM0G3507, MSPM0G3519, MSPM0G5187, MSPM33C32, F29H85, AM13E2, AM263 Detect DC arc faults from current waveforms for electrical safety
ecg_classification ecg_classification MSPM0G3507, MSPM0G5187, MSPM0G3519 Classify normal vs anomalous heartbeats from ECG signals
fall_detection_classification classification MSPM0G5187 Detect and classify Human Fall vs Activities of Daily Living (ADL)
gearbox_fault_detection classification MSPM0G3507, MSPM0G3519, MSPM0G5187 Classify gearbox operating conditions (healthy vs broken tooth) from vibration data
blower_imbalance motor_fault F280013, F280015, F28003, F28004, F2837, F28P55, F28P65, MSPM0G3507, MSPM0G3519, MSPM0G5187, MSPM33C32, F29H85, AM13E2, AM263 Detect blade imbalance in HVAC blowers using 3-phase motor currents
fan_blade_fault_classification motor_fault F280013, F280015, F28003, F28004, F2837, F28P55, F28P65, MSPM0G3507, MSPM0G3519, MSPM0G5187, MSPM33C32, F29H85, AM13E2, AM263, CC1312, CC1314, CC1352, CC1354, CC2755, CC35X1 Detect faults in BLDC fans from accelerometer data
motor_bearing_fault motor_fault F280013, F280015, F28003, F28004, F2837, F28P55, F28P65, MSPM0G3507, MSPM0G3519, MSPM0G5187, MSPM33C32, F29H85, AM13E2, AM263 Classify 5 bearing fault types + normal operation from vibration data
grid_fault_detection classification F280013, F280015, F28003, F28004, F2837, F28P55, F28P65, MSPM0G3507, MSPM0G3519, MSPM0G5187, MSPM33C32, F29H85, AM13E2, AM263 Detect electrical grid faults from sensor data
mosfet_temp_prediction regression F280013, F280015, F28003, F28004, F2837, F28P55, F28P65, MSPM0G3507, MSPM0G3519, MSPM0G5187, MSPM33C32, F29H85, AM13E2, AM263 Predict MOSFET temperature from electrical parameters
pir_detection pir_detection CC2755, CC1312, CC1352, CC1354, CC35X1, MSPM0G5187, MSPM0G3507, MSPM0G3519, MSPM33C32 Detect presence/motion using PIR sensor data
MNIST_image_classification image_classification MSPM0G3507, MSPM0G3519, MSPM0G5187, MSPM33C32 Handwritten digit recognition (MNIST dataset)

Summary by Task Type:

  • Generic Timeseries Tasks (22 examples): Support all target devices and can be adapted to your custom datasets
    • Classification: 8 examples (1 base + 7 real-world applications)
    • Regression: 5 examples (1 base + 4 real-world applications)
    • Forecasting: 3 examples (1 base + 2 real-world applications)
    • Anomaly Detection: 6 examples (1 base + 5 application variants)
  • Application-Specific Tasks (12 examples): arc_fault (2), motor_fault (3), grid_fault_detection (1), gearbox_fault_detection (1), mosfet_temp_prediction (1), pir_detection (1), ecg_classification (1), image_classification (1), fall_detection_classification (1)

Detailed Examples by Task Type

Classification Examples

No. Example Data Type Description
1 dc_arc_fault Univariate Detect DC arc faults from current waveforms for electrical safety.
2 ac_arc_fault Univariate Detect AC arc faults in electrical systems.
3 motor_bearing_fault Multivariate Classify 5 bearing fault types + normal operation from vibration data.
4 blower_imbalance Multivariate Detect blade imbalance in HVAC blowers using 3-phase motor currents.
5 fan_blade_fault_classification Multivariate Detect faults in BLDC fans from accelerometer data.
6 gearbox_fault_detection Multivariate Classify gearbox operating conditions (healthy vs broken tooth) from vibration.
7 electrical_fault Multivariate Classify transmission line faults using voltage and current (2-class and 6-class variants).
8 grid_stability Multivariate Predict power grid stability from node parameters.
9 gas_sensor Multivariate Identify gas type and concentration from sensor array data.
10 branched_model_parameters Multivariate Human Activity Recognition from accelerometer/gyroscope data.
11 ecg_classification Multivariate Classify normal vs anomalous heartbeats from ECG signals.
12 nilm_appliance_usage_classification Multivariate Non-Intrusive Load Monitoring - identify active appliances.
13 PLAID_nilm_classification Multivariate Appliance identification using the PLAID dataset.
14 pir_detection Multivariate Detect presence/motion using PIR sensor data.
15 fall_detection_classification Multivariate Detect and classify Human Fall vs Activities of Daily Living (ADL).

Regression Examples

No. Example Data Type Description
1 torque_measurement_regression Multivariate Predict PMSM motor torque from current measurements.
2 induction_motor_speed_prediction Multivariate Predict induction motor speed from electrical signals.
3 reg_washing_machine Multivariate Predict washing machine load weight.

Forecasting Examples

No. Example Data Type Description
1 forecasting_pmsm_rotor Multivariate Forecast PMSM rotor winding temperature.
2 hvac_indoor_temp_forecast Multivariate Predict indoor temperature for HVAC control.

Anomaly Detection Examples

No. Example Data Type Description
1 dc_arc_fault (DSI) Univariate Detect anomalous DC arc patterns using autoencoder (DSI dataset).
2 dc_arc_fault (DSK) Univariate Detect anomalous DC arc patterns using autoencoder (DSK dataset).
3 ecg_classification Multivariate Detect anomalous heartbeat patterns from ECG signals.
4 fan_blade_fault_classification Multivariate Detect anomalous fan blade behavior from accelerometer data.
5 motor_bearing_fault Multivariate Detect anomalous bearing behavior from vibration data.

Image Classification Examples

No. Example Data Type Description
1 MNIST_image_classification Image Handwritten digit recognition (MNIST dataset).

Available Models

Models are organized by task type. The NPU column indicates hardware acceleration support on TI devices with NPU (F28P55, AM13E2, MSPM0G5187).

NPU-optimized models follow specific layer constraints for hardware acceleration:

  • All channels are multiples of 4 (m4)
  • Kernel heights ≤ 7 for GCONV layers
  • MaxPool kernels ≤ 4
  • FC layer inputs ≥ 16 features (8-bit) or ≥ 8 features (4-bit)

For detailed guidelines, see NPU Configuration Guidelines.

When to use NPU-optimized models:

  • Target device has NPU (F28P55, AM13E2, MSPM0G5187)
  • You need maximum inference speed
  • Standard models show "fallback to software" warnings during compilation

Classification Models

Model Name Parameters Architecture NPU Description
CLS_100_NPU ~100 CNN Yes Ultra-compact model
CLS_500_NPU ~500 CNN Yes Compact model
CLS_1k_NPU ~1K CNN Yes Lightweight 2-layer CNN
CLS_1.2k_NPU ~1.2K CNN Yes Compact model for ultra-low power devices
CLS_1.5k_NPU ~1.5K CNN Yes 3-layer model with balanced performance
CLS_1.9k_NPU ~1.9K CNN Yes Efficient 3-layer model
CLS_2k_NPU ~2K CNN Yes 2-layer model
CLS_2.8k_NPU ~2.8K CNN Yes Improved accuracy with compact footprint
CLS_3.1k_NPU ~3.1K CNN Yes Higher accuracy model
CLS_ResAdd_3k ~3K ResNet (Add) No Residual connections with addition
CLS_ResCat_3k ~3K ResNet (Cat) No Residual connections with concatenation
CLS_3.9k_NPU ~3.9K CNN Yes Advanced 3-layer model
CLS_4k_NPU ~4K CNN Yes Balanced model
CLS_4.2k_NPU ~4.2K CNN Yes Optimized 4-layer model
CLS_5k_NPU ~5K CNN Yes Mid-range model
CLS_6k_NPU ~6K CNN (DW-Sep) Yes Depthwise separable
CLS_8k_NPU ~8K CNN (DW-Sep) Yes Depthwise separable
CLS_13k_NPU ~13K CNN Yes Higher capacity
CLS_20k_NPU ~20K CNN Yes High capacity
CLS_40k_NPU ~40K CNN Yes Advanced model for complex tasks
CLS_55k_NPU ~55K CNN Yes Maximum accuracy
ArcFault_model_200_t ~200 Specialized No Arc fault detection
ArcFault_model_300_t ~300 Specialized No Arc fault with more capacity
ArcFault_model_700_t ~700 Specialized No Arc fault medium model
ArcFault_model_1400_t ~1.4K Specialized No Arc fault high accuracy
GearboxFault_model_1.2k_t ~1.2K CNN Yes Gearbox fault detection
GearboxFault_model_1.5k_t ~1.5K CNN Yes Gearbox fault with more capacity
MotorFault_model_1_t Varies Specialized No Motor bearing fault detection
MotorFault_model_2_t Varies Specialized No Motor fault variant 2
MotorFault_model_3_t Varies Specialized No Motor fault variant 3
FanImbalance_model_1_t Varies Specialized No Fan blade imbalance detection
FanImbalance_model_2_t Varies Specialized No Fan imbalance variant 2
FanImbalance_model_3_t Varies Specialized No Fan imbalance variant 3
PIRDetection_model_1_t Varies Specialized No PIR-based presence detection

Regression Models

Model Name Parameters Architecture NPU Description
REGR_500_NPU ~500 CNN Yes Compact regression
REGR_1k ~1K CNN No Lightweight regression model
REGR_2k_NPU ~2K CNN Yes 2-layer model
REGR_3k ~3K MLP No 4-layer fully connected network
REGR_4k ~4K CNN No 2 Conv+BN+ReLU + Linear
REGR_6k_NPU ~6K CNN (DW-Sep) Yes Depthwise separable convolutions
REGR_8k_NPU ~8K CNN Yes 3-layer model
REGR_10k ~10K CNN No 3 Conv+BN+ReLU + 2 Linear
REGR_13k ~13K CNN No High capacity regression
REGR_20k_NPU ~20K CNN Yes High capacity with MaxPool

Anomaly Detection Models

Note: For NPU models, encoder convolutions are NPU-accelerated but decoder upsampling falls back to CPU.

Model Name Parameters Architecture NPU Description
AD_500_NPU ~500 CNN AE Yes 2-layer autoencoder
AD_1k ~1K Autoencoder No Compact autoencoder
AD_2k_NPU ~2K CNN AE Yes 2-layer autoencoder
AD_4k ~4K Autoencoder No 3-layer CNN autoencoder
AD_6k_NPU ~6K CNN AE (DW-Sep) Yes Depthwise separable encoder
AD_8k_NPU ~8K CNN AE Yes 3-layer autoencoder
AD_10k_NPU ~10K CNN AE Yes 3-layer autoencoder
AD_16k ~16K Autoencoder No 4-layer CNN autoencoder
AD_17k ~17K Autoencoder No Fan blade anomaly detection
AD_20k_NPU ~20K CNN AE Yes High capacity autoencoder
AD_Linear Varies Linear AE No 3-layer deep linear autoencoder
Ondevice_Trainable_AD_Linear Varies Linear AE No On-device trainable variant

Forecasting Models

Note: LSTM models are not NPU-supported.

Model Name Parameters Architecture NPU Description
FCST_500_NPU ~500 CNN Yes Compact forecasting
FCST_1k_NPU ~1K CNN Yes 2-layer model
FCST_2k_NPU ~2K CNN Yes 2-layer model
FCST_3k ~3K MLP No 4-layer fully connected
FCST_4k_NPU ~4K CNN Yes 3-layer model
FCST_6k_NPU ~6K CNN (DW-Sep) Yes Depthwise separable convolutions
FCST_8k_NPU ~8K CNN Yes 3-layer model
FCST_10k_NPU ~10K CNN Yes 3-layer model
FCST_13k ~13K CNN No 2 Conv+BN+ReLU + Linear
FCST_20k_NPU ~20K CNN Yes High capacity with MaxPool
FCST_LSTM8 Varies LSTM No Single LSTM (hidden=8) + Linear
FCST_LSTM10 Varies LSTM No Single LSTM (hidden=10) + Linear

Image Classification Models

Model Name Parameters Architecture NPU Description
Lenet5 ~60K LeNet-5 No Classic CNN for image classification

Adding New Models

Want to add your own model? See the comprehensive guide: ADDING_NEW_MODELS.md

Key steps:

  1. Add model class to tinyml_modelzoo/models/
  2. Add class name to the file's __all__ list
  3. (Optional) Add device performance info to device_info/run_info.py
  4. (Optional) Add model description to model_descriptions/ for GUI integration

No changes required in tinyml-tinyverse or tinyml-modelmaker!

Additional Resources

License

BSD 3-Clause License. See LICENSE for details.

Questions or Issues? Open an issue on GitHub.