LogoLogo
HomeAPI & SDKsProjectsForumStudio
  • Getting started
    • For beginners
    • For ML practitioners
    • For embedded engineers
  • Frequently asked questions (FAQ)
  • Tutorials
    • End-to-end tutorials
      • Computer vision
        • Image classification
        • Object detection
          • Object detection with bounding boxes
          • Detect objects with centroid (FOMO)
        • Visual anomaly detection
        • Visual regression
      • Audio
        • Sound recognition
        • Keyword spotting
      • Time-series
        • Motion recognition + anomaly detection
        • Regression + anomaly detection
        • HR/HRV
        • Environmental (Sensor fusion)
    • Data
      • Data ingestion
        • Collecting image data from the Studio
        • Collecting image data with your mobile phone
        • Collecting image data with the OpenMV Cam H7 Plus
        • Using the Edge Impulse Python SDK to upload and download data
        • Trigger connected board data sampling
        • Ingest multi-labeled data using the API
      • Synthetic data
        • Generate audio datasets using Eleven Labs
        • Generate image datasets using Dall-E
        • Generate keyword spotting datasets using Google TTS
        • Generate physics simulation datasets using PyBullet
        • Generate timeseries data with MATLAB
      • Labeling
        • Label audio data using your existing models
        • Label image data using GPT-4o
      • Edge Impulse Datasets
    • Feature extraction
      • Building custom processing blocks
      • Sensor fusion using embeddings
    • Machine learning
      • Classification with multiple 2D input features
      • Visualize neural networks decisions with Grad-CAM
      • Sensor fusion using embeddings
      • FOMO self-attention
    • Inferencing & post-processing
      • Count objects using FOMO
      • Continuous audio sampling
      • Multi-impulse (C++)
      • Multi-impulse (Python)
    • Lifecycle management
      • CI/CD with GitHub Actions
      • Data aquisition from S3 object store - Golioth on AI
      • OTA model updates
        • with Arduino IDE (for ESP32)
        • with Arduino IoT Cloud
        • with Blues Wireless
        • with Docker on Allxon
        • with Docker on Balena
        • with Docker on NVIDIA Jetson
        • with Espressif IDF
        • with Nordic Thingy53 and the Edge Impulse app
        • with Particle Workbench
        • with Zephyr on Golioth
    • API examples
      • Customize the EON Tuner
      • Ingest multi-labeled data using the API
      • Python API bindings example
      • Running jobs using the API
      • Trigger connected board data sampling
    • Python SDK examples
      • Using the Edge Impulse Python SDK to run EON Tuner
      • Using the Edge Impulse Python SDK to upload and download data
      • Using the Edge Impulse Python SDK with Hugging Face
      • Using the Edge Impulse Python SDK with SageMaker Studio
      • Using the Edge Impulse Python SDK with TensorFlow and Keras
      • Using the Edge Impulse Python SDK with Weights & Biases
    • Expert network projects
  • Edge Impulse Studio
    • Organization hub
      • Users
      • Data campaigns
      • Data
        • Cloud data storage
      • Data pipelines
      • Data transformation
        • Transformation blocks
      • Upload portals
      • Custom blocks
        • Custom AI labeling blocks
        • Custom deployment blocks
        • Custom learning blocks
        • Custom processing blocks
        • Custom synthetic data blocks
        • Custom transformation blocks
      • Health reference design
        • Synchronizing clinical data with a bucket
        • Validating clinical data
        • Querying clinical data
        • Transforming clinical data
    • Project dashboard
      • Select AI hardware
    • Devices
    • Data acquisition
      • Uploader
      • Data explorer
      • Data sources
      • Synthetic data
      • Labeling queue
      • AI labeling
      • CSV Wizard (time-series)
      • Multi-label (time-series)
      • Tabular data (pre-processed & non-time-series)
      • Metadata
      • Auto-labeler | deprecated
    • Impulses
    • EON Tuner
      • Search space
    • Processing blocks
      • Audio MFCC
      • Audio MFE
      • Audio Syntiant
      • Flatten
      • HR/HRV features
      • Image
      • IMU Syntiant
      • Raw data
      • Spectral features
      • Spectrogram
      • Custom processing blocks
      • Feature explorer
    • Learning blocks
      • Anomaly detection (GMM)
      • Anomaly detection (K-means)
      • Classification
      • Classical ML
      • Object detection
        • MobileNetV2 SSD FPN
        • FOMO: Object detection for constrained devices
      • Object tracking
      • Regression
      • Transfer learning (images)
      • Transfer learning (keyword spotting)
      • Visual anomaly detection (FOMO-AD)
      • Custom learning blocks
      • Expert mode
      • NVIDIA TAO | deprecated
    • Retrain model
    • Live classification
    • Model testing
    • Performance calibration
    • Deployment
      • EON Compiler
      • Custom deployment blocks
    • Versioning
    • Bring your own model (BYOM)
    • File specifications
      • deployment-metadata.json
      • ei-metadata.json
      • ids.json
      • parameters.json
      • sample_id_details.json
      • train_input.json
  • Tools
    • API and SDK references
    • Edge Impulse CLI
      • Installation
      • Serial daemon
      • Uploader
      • Data forwarder
      • Impulse runner
      • Blocks
      • Himax flash tool
    • Edge Impulse for Linux
      • Linux Node.js SDK
      • Linux Go SDK
      • Linux C++ SDK
      • Linux Python SDK
      • Flex delegates
      • Rust Library
    • Rust Library
    • Edge Impulse Python SDK
  • Run inference
    • C++ library
      • As a generic C++ library
      • On Android
      • On your desktop computer
      • On your Alif Ensemble Series Device
      • On your Espressif ESP-EYE (ESP32) development board
      • On your Himax WE-I Plus
      • On your Raspberry Pi Pico (RP2040) development board
      • On your SiLabs Thunderboard Sense 2
      • On your Spresense by Sony development board
      • On your Syntiant TinyML Board
      • On your TI LaunchPad using GCC and the SimpleLink SDK
      • On your Zephyr-based Nordic Semiconductor development board
    • Arm Keil MDK CMSIS-PACK
    • Arduino library
      • Arduino IDE 1.18
    • Cube.MX CMSIS-PACK
    • Docker container
    • DRP-AI library
      • DRP-AI on your Renesas development board
      • DRP-AI TVM i8 on Renesas RZ/V2H
    • IAR library
    • Linux EIM executable
    • OpenMV
    • Particle library
    • Qualcomm IM SDK GStreamer
    • WebAssembly
      • Through WebAssembly (Node.js)
      • Through WebAssembly (browser)
    • Edge Impulse firmwares
    • Hardware specific tutorials
      • Image classification - Sony Spresense
      • Audio event detection with Particle boards
      • Motion recognition - Particle - Photon 2 & Boron
      • Motion recognition - RASynBoard
      • Motion recognition - Syntiant
      • Object detection - SiLabs xG24 Dev Kit
      • Sound recognition - TI LaunchXL
      • Keyword spotting - TI LaunchXL
      • Keyword spotting - Syntiant - RC Commands
      • Running NVIDIA TAO models on the Renesas RA8D1
      • Two cameras, two models - running multiple object detection models on the RZ/V2L
  • Edge AI Hardware
    • Overview
    • Production-ready
      • Advantech ICAM-540
      • Seeed SenseCAP A1101
      • Industry reference design - BrickML
    • MCU
      • Ambiq Apollo4 family of SoCs
      • Ambiq Apollo510
      • Arducam Pico4ML TinyML Dev Kit
      • Arduino Nano 33 BLE Sense
      • Arduino Nicla Sense ME
      • Arduino Nicla Vision
      • Arduino Portenta H7
      • Blues Wireless Swan
      • Espressif ESP-EYE
      • Himax WE-I Plus
      • Infineon CY8CKIT-062-BLE Pioneer Kit
      • Infineon CY8CKIT-062S2 Pioneer Kit
      • Nordic Semi nRF52840 DK
      • Nordic Semi nRF5340 DK
      • Nordic Semi nRF9160 DK
      • Nordic Semi nRF9161 DK
      • Nordic Semi nRF9151 DK
      • Nordic Semi nRF7002 DK
      • Nordic Semi Thingy:53
      • Nordic Semi Thingy:91
      • Open MV Cam H7 Plus
      • Particle Photon 2
      • Particle Boron
      • RAKwireless WisBlock
      • Raspberry Pi RP2040
      • Renesas CK-RA6M5 Cloud Kit
      • Renesas EK-RA8D1
      • Seeed Wio Terminal
      • Seeed XIAO nRF52840 Sense
      • Seeed XIAO ESP32 S3 Sense
      • SiLabs Thunderboard Sense 2
      • Sony's Spresense
      • ST B-L475E-IOT01A
      • TI CC1352P Launchpad
    • MCU + AI accelerators
      • Alif Ensemble
      • Arduino Nicla Voice
      • Avnet RASynBoard
      • Seeed Grove - Vision AI Module
      • Seeed Grove Vision AI Module V2 (WiseEye2)
      • Himax WiseEye2 Module and ISM Devboard
      • SiLabs xG24 Dev Kit
      • STMicroelectronics STM32N6570-DK
      • Synaptics Katana EVK
      • Syntiant Tiny ML Board
    • CPU
      • macOS
      • Linux x86_64
      • Raspberry Pi 4
      • Raspberry Pi 5
      • Texas Instruments SK-AM62
      • Microchip SAMA7G54
      • Renesas RZ/G2L
    • CPU + AI accelerators
      • AVNET RZBoard V2L
      • BrainChip AKD1000
      • i.MX 8M Plus EVK
      • Digi ConnectCore 93 Development Kit
      • MemryX MX3
      • MistyWest MistySOM RZ/V2L
      • Qualcomm Dragonwing RB3 Gen 2 Dev Kit
      • Renesas RZ/V2L
      • Renesas RZ/V2H
      • IMDT RZ/V2H
      • Texas Instruments SK-TDA4VM
      • Texas Instruments SK-AM62A-LP
      • Texas Instruments SK-AM68A
      • Thundercomm Rubik Pi 3
    • GPU
      • Advantech ICAM-540
      • NVIDIA Jetson
      • Seeed reComputer Jetson
    • Mobile phone
    • Porting guide
  • Integrations
    • Arduino Machine Learning Tools
    • AWS IoT Greengrass
    • Embedded IDEs - Open-CMSIS
    • NVIDIA Omniverse
    • Scailable
    • Weights & Biases
  • Tips & Tricks
    • Combining impulses
    • Increasing model performance
    • Optimizing compute time
    • Inference performance metrics
  • Concepts
    • Glossary
    • Course: Edge AI Fundamentals
      • Introduction to edge AI
      • What is edge computing?
      • What is machine learning (ML)?
      • What is edge AI?
      • How to choose an edge AI device
      • Edge AI lifecycle
      • What is edge MLOps?
      • What is Edge Impulse?
      • Case study: Izoelektro smart grid monitoring
      • Test and certification
    • Data engineering
      • Audio feature extraction
      • Motion feature extraction
    • Machine learning
      • Data augmentation
      • Evaluation metrics
      • Neural networks
        • Layers
        • Activation functions
        • Loss functions
        • Optimizers
          • Learned optimizer (VeLO)
        • Epochs
    • What is embedded ML, anyway?
    • What is edge machine learning (edge ML)?
Powered by GitBook
On this page
  • How does this 🪄 work?
  • Heat maps
  • Training on centroids
  • Flexible and very, very fast
  • How to get started?
  • Expert mode tips
  • Object weighting
  • MobileNet cut point
  • FOMO classifier capacity
  • Performance and Minimum Requirements

Was this helpful?

Export as PDF
  1. Edge Impulse Studio
  2. Learning blocks
  3. Object detection

FOMO: Object detection for constrained devices

PreviousMobileNetV2 SSD FPNNextObject tracking

Last updated 2 months ago

Was this helpful?

Edge Impulse FOMO (Faster Objects, More Objects) is a novel machine learning algorithm that brings object detection to highly constrained devices which lets you count multiple objects and find their location in an image in real-time using up to 30x less processing power and memory than MobileNet SSD or YOLOv5.

Tutorials

Want to see FOMO in action? Check out our tutorial.

For example, FOMO lets you do 60 fps object detection on a Raspberry Pi 4:

And here's FOMO doing 30 fps object detection on an Arduino Nicla Vision (Cortex-M7 MCU), using 245K RAM.

How does this 🪄 work?

So how does that work? First, a small primer. Let's say you want to detect whether you see a face in front of your sensor. You can approach this in two ways. You can train a simple binary classifier, which says either "face" or "no face", or you can train a complex object detection model which tells you "I see a face at this x, y point and of this size". Object detection is thus great when you need to know the exact location of something, or if you want to count multiple things (the simple classifier cannot do that) - but it's computationally much more intensive, and you typically need much more data for it.

The design goal for FOMO was to get the best of both worlds: the computational power required for simple image classification, but with the additional information on location and object count that object detection gives us.

Heat maps

The first thing to realize is that while the output of the image classifier is "face" / "no face" (and thus no locality is preserved in the outcome) the underlying neural network architecture consists of a number of convolutional layers. A way to think about these layers is that every layer creates a diffused lower-resolution image of the previous layer. E.g. if you have a 16x16 image the width/height of the layers may be:

  1. 16x16

  2. 4x4

  3. 1x1

Each 'pixel' in the second layer maps roughly to a 4x4 block of pixels in the input layer, and the interesting part is that locality is somewhat preserved. The 'pixel' in layer 2 at (0, 0) will roughly map back to the top left corner of the input image. The deeper you go in a normal image classification network, the less of this locality (or "receptive field") is preserved until you finally have just 1 outcome.

FOMO uses the same architecture, but cuts off the last layers of a standard image classification model and replaces this layer with a per-region class probability map (e.g. a 4x4 map in the example above). It then has a custom loss function which forces the network to fully preserve the locality in the final layer. This essentially gives you a heatmap of where the objects are.

The resolution of the heat map is determined by where you cut off the layers of the network. For the FOMO model trained above (on the beer bottles) we do this when the size of the heat map is 8x smaller than the input image (input image of 160x160 will yield a 20x20 heat map), but this is configurable. When you set this to 1:1 this actually gives you pixel-level segmentation and the ability to count a lot of small objects.

Training on centroids

A difference between FOMO and other object detection algorithms is that it does not output bounding boxes, but it's easy to go from heat map to bounding boxes. Just draw a box around a highlighted area.

However, when working with early customers we realized that bounding boxes are merely an implementation detail of other object detection networks, and are not a typical requirement. Very often the size of objects is not important as cameras are in fixed locations (and objects thus fixed size), but rather you just want the location and the count of objects.

Thus, we now train on the centroids of objects. This makes it much easier to count objects that are close (every activation in the heat map is an object), and the convolutional nature of the neural network ensures we look around the centroid for the object anyway.

A downside of the heat map is that each cell acts as its own classifier. E.g. if your classes are "lamp", "plant" and "background" each cell will be either lamp, plant, or background. It's thus not possible to detect objects with overlapping centroids. You can see this in the Raspberry Pi 4 video above at 00:18 where the beer bottles are too close together. This can be solved by using a higher resolution heat map.

Flexible and very, very fast

A really cool benefit of FOMO is that it's fully convolutional. If you set an image:heat map factor of 8 you can throw in a 96x96 image (outputs 12x12 heat map), a 320x320 image (outputs 40x40 heat map), or even a 1024x1024 image (outputs 128x128 heat map). This makes FOMO incredibly flexible, and useful even if you have very large images that need to be analyzed (e.g. in fault detection where the faults might be very, very small). You can even train on smaller patches, and then scale up during inference.

Additionally FOMO is compatible with any MobileNetV2 model. Depending on where the model needs to run you can pick a model with a higher or lower alpha, and transfer learning also works (although you need to train your base models specifically with FOMO in mind). This makes it easy for end customers to use their existing models and fine-tune them with FOMO to also add locality (e.g. we have customers with large transfer learning models for wildlife detection).

Together this gives FOMO the capabilities to scale from the smallest microcontrollers all the way to full gateways or GPUs. Just some numbers:

  1. The video on the top classifies 60 times / second on a stock Raspberry Pi 4 (160x160 grayscale input, MobileNetV2 0.1 alpha). This is 20x faster than MobileNet SSD which does ~3 frames/second.

  2. The second video on the top classifies 30 times / second on an Arduino Nicla Vision board with a Cortex-M7 MCU running at 480MHz) in ~240K of RAM (96x96 grayscale input, MobileNetV2 0.35 alpha).

  3. The smallest version of FOMO (96x96 grayscale input, MobileNetV2 0.1 alpha) runs under 100KB RAM and ~10 fps. on a Cortex-M4F at 80MHz. [1]

How to get started?

To build your first FOMO models:

  1. Create a new project in Edge Impulse.

  2. Make sure to set your labeling method to 'Bounding boxes (object detection)'.

  3. Add an 'Object Detection (Images)' block to your impulse.

  4. Under Images, select 'Grayscale'

  5. Under Object detection, select 'Choose a different model' and select one of the FOMO models.

  6. Make sure to lower the learning rate to 0.001 to start.

Expert mode tips

Additional configuration for FOMO can be accessed via expert mode.

Object weighting

FOMO is sensitive to the ratio of objects to background cells in the labelled data. By default the configuration is to weight object output cells x100 in the loss function, object_weight=100, as a way of balancing what is usually a majority of background. This value was chosen as a sweet spot for a number of example use cases. In scenarios where the objects to detect are relatively rare this value can be increased, e.g. to 1000, to have the model focus even more on object detection (at the expense of potentially more false detections).

MobileNet cut point

Choosing a different cut_point results in a different spatial reduction; e.g. if we cut higher at block_3_expand_relu FOMO will instead only do a spatial reduction of 1/4 (i.e. a 96x96 input results in a 24x24output)

Note though; this means taking much less of the MobileNet backbone and results in a model with only 1/2 the params. Switching to a higher alpha may counteract this parameter reduction. Later FOMO releases will counter this parameter reduction with a UNet style architecture.

FOMO classifier capacity

FOMO can be thought of logically as the first section of MobileNetV2 followed by a standard classifier where the classifier is applied in a fully convolutional fashion.

In the default configuration this FOMO classifier is equivalent to a single dense layer with 32 nodes followed by a classifier with num_classes outputs.

For a three way classifier, using the default cut point, the result is a classifier head with ~3200 parameters.

 LAYER                          SHAPE                NUMBER OF PARAMETERS
 block_6_expand_relu (ReLU)     (None, 20, 20, 96)   0
 head (Conv2D)                  (None, 20, 20, 32)   3104
 logits (Conv2D)                (None, 20, 20, 3)    99

We have the option of increasing the capacity of this classifier head by either 1) increasing the number of filters in the Conv2D layer, 2) adding additional layers or 3) doing both.

For example we might change the number of filters from 32 to 16, as well as adding another convolutional layer, as follows.

 LAYER                          SHAPE                NUMBER OF PARAMETERS
 block_6_expand_relu (ReLU)     (None, 20, 20, 96)   0
 head_1 (Conv2D)                (None, 20, 20, 16)   1552
 head_2 (Conv2D)                (None, 20, 20, 16)   272
 logits (Conv2D)                (None, 20, 20, 3)    51

For some problems an additional layer can improve performance, and in this case actually uses less parameters. It can though potentially take longer to train and require more data. In future releases the tuning of this aspect of FOMO can be handled by the EON Tuner.

Performance and Minimum Requirements

Just like the rest of our Neural Network-based learning blocks, FOMO is delivered as a set of basic math routines free of runtime dependencies. This means that there are virtually no limitations to running FOMO, other than:

  1. Making sure the model itself can fit into the target's memory (flash/RAM), and

  2. making sure the target also has enough memory to hold the image buffer (flash/RAM)in addition to your application logic

In all, we have seen buffer, model and app logic (including wireless stack) fit in as little as 200KB for 64x64 pixel images. But we would definitely recommend a target with at least 512KB so that you can take advantage of larger image sizes and a wider range of model optimizations.

With regards to latency, the speed of the target will determine the maximum number of frames that can be processed in a given interval (fps). This will of course be influenced by any other tasks the CPU may need to complete, but we have consistently seen MCUs running @ 80MHz complete a full pass on a 64x64 pixel image in under one second, which should translate to just under 1fps once you add the rest of your app logic. Keep in mind that frame throughput can increase dramatically at higher speeds or when tensor acceleration is available. We have measured 40-60 fps consistently on a Raspberry Pi 4 and ~15 fps on unaccelerated 480MHz targets. The table below summarizes this trade-off:

Requirement

Minimum

Recommended

Memory Footprint (RAM)

256 KB 64x64 pixels (B&W, buffer included)

≥ 512 KB 96x96 pixels (B&W, buffer Included)

Latency (100% load)

80 MHz < 1 fps

> 80 MHz + acceleration ~15 fps @ 480MHz 40-60fps in RPi4

You can find the complete Edge Impulse project with the beers vs. cans model, including all data and configuration here: .

During Edge Impulse Imagine we demonstrated a FOMO model running on a doing 14 frames per second on a DSP (). This model ran in under 150KB of RAM (96x96 grayscale input, MobileNetV2 0.1 alpha). [1]

[1] Models compiled using .

Collect and prepare your dataset as in .

FOMO is currently compatible with all that have a camera, and with Edge Impulse for Linux (any client). Of course, you can export your model as a C++ Library and integrate it as usual on any device or development board, the output format of models is compatible with normal object detection models; and our SDK runs on almost anything under the sun (see for an overview) from RTOS's to bare-metal to special accelerators and GPUs.

FOMO uses as a base model for its trunk and by default does a spatial reduction of 1/8th from input to output (e.g. a 96x96 input results in a 12x12 output). This is implemented by cutting MobileNet off at the intermediate layer block_6_expand_relu

https://studio.edgeimpulse.com/public/89078/latest
Himax WE-I Plus
video
EON Compiler
object detection
fully-supported development boards
Running your impulse locally
MobileNetV2
Detect objects with centroids (FOMO)
Image classification vs object detection
From input image to heat map (cup in red, lamp in green)
Here's a former iteration of the FOMO approach used to count individual bees (heat map 2x smaller than the input size).
From heat map to bounding boxes
Training on the centroids of beer bottles. On top the source labels, at the bottom the inference result.
Selecting a FOMO model in Edge Impulse
Accessing expert mode
MobileNetV2 cut point
FOMO uses a convolutional classifier.
Adding an additional layer to the classifier of FOMO