Object detection

An object detection model can identify multiple objects and their location in an image.

With the Coral Edge TPU™, you can run an object detection model directly on your device, using real-time video, at over 100 frames per second. You can even run multiple detection models concurrently on one Edge TPU, while maintaining a high frame rate.

This page provides several trained models that are compiled for the Edge TPU, example code to run them, plus information about how to train your own model with TensorFlow.

Trained models link

These models are trained and compiled for the Edge TPU.

Notice: These are not production-quality models; they are for demonstration purposes only.
Model name Detections/Dataset Input size TF ver. Latency 1 mAP 2 Micro 3 Model size Downloads

SSD MobileNet V1

90 objects
COCO

300x300x3 1 6.5 ms 21.5% Yes 7.0 MB

SSD/FPN MobileNet V1
New

90 objects
COCO

640x640x3 2 229.4 ms 31.1% No 37.7 MB

SSD MobileNet V2

90 objects
COCO

300x300x3 1 7.3 ms 25.6% Yes 6.6 MB

SSD MobileNet V2
New

90 objects
COCO

300x300x3 2 7.6 ms 22.4% Yes 6.7 MB

SSD MobileNet V2

Faces

320x320x3 1 5.2 ms N/A Yes 6.7 MB

SSDLite MobileDet

90 objects
COCO

320x320x3 1 9.1 ms 32.9% Yes 5.1 MB

EfficientDet-Lite0
New

90 objects
COCO

320x320x3 2 37.4 ms 30.4% No 5.7 MB

EfficientDet-Lite1
New

90 objects
COCO

384x384x3 2 56.3 ms 34.3% No 7.6 MB

EfficientDet-Lite2
New

90 objects
COCO

448x448x3 2 104.6 ms 36.0% No 10.2 MB

EfficientDet-Lite3
New

90 objects
COCO

512x512x3 2 107.6 ms 39.4% No 14.4 MB

EfficientDet-Lite3x*
New

90 objects
COCO

640x640x3 2 197.0 ms 43.9% No 20.6 MB

1 Latency is the time to perform one inference, as measured with a Coral USB Accelerator on a desktop CPU. Latency varies between systems and is primarily intended for comparison between models. For more comparisons, see the Performance Benchmarks.

2 mAP is the "mean average precision," as specified by the COCO evaluation metrics. Our evaluation uses a subset of the COCO17 dataset.

3 Indicates compatibility with the Dev Board Micro. Some models are not compatible because they require a CPU-bound op that is not supported by TensorFlow Lite for Microcontrollers or they require more memory than available on the board. (All models are compatible with all other Coral boards.)

* EfficientDet-Lite3x is not compatible with Edge TPUs over USB; it can be used only with PCIe-based devices. Our benchmarks for EfficientDet-Lite3x come from a desktop system paired with the Asus AI Accelerator—your results will vary.

Example code link

Basic object detection

An example that performs object detection with a photo and draws a square around each object. Also works with face detection models.

Languages: Python

videocam

Image recognition with video

Multiple examples showing how to stream images from a camera and run classification or detection models with the TensorFlow Lite API. Each example uses a different camera library, such as GStreamer, OpenCV, PyGame, and PiCamera.

Languages: Python

videocam

Object tracking with video

This example takes a camera feed and tracks each uniquely identified object, assigning each object with a persistent ID. The example detection script allows you to specify the tracker program you want to use (the Sort tracker is included).

Languages: Python