Image classification

An image classification model can label an image with known objects and provide a confidence score.

With the Coral Edge TPU™, you can run an image classification model directly on your device, using real-time video at almost 400 frames per second. You can even run additional models concurrently on the same 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

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	Depth mul.	TF ver.	Latency ¹	Accuracy	Micro ²	Model size	Downloads
EfficientNet-EdgeTpu (L)^*	1,000 objects ILSVRC2012	300x300x3	N/A	1	21.3 ms	Top-1: 81.2% Top-5: 95.1%	Yes	12.8 MB	Edge TPU model, CPU model, Labels file, All model files
EfficientNet-EdgeTpu (M)^*	1,000 objects ILSVRC2012	240x240x3	N/A	1	7.3 ms	Top-1: 80.1% Top-5: 94.5%	Yes	8.7 MB	Edge TPU model, CPU model, Labels file, All model files
EfficientNet-EdgeTpu (S)^*	1,000 objects ILSVRC2012	224x224x3	N/A	1	5.0 ms	Top-1: 78.9% Top-5: 94.7%	Yes	6.8 MB	Edge TPU model, CPU model, Labels file, All model files
Inception V1	1,000 objects ILSVRC2012	224x224x3	N/A	1	3.4 ms	Top-1: 71.9% Top-5: 92.0%	Yes	7.0 MB	Edge TPU model, CPU model, Labels file, All model files
Inception V3	1,000 objects ILSVRC2012	224x224x3	N/A	1	13.4 ms	Top-1: 75.4% Top-5: 93.2%	Yes	12.0 MB	Edge TPU model, CPU model, Labels file, All model files
Inception V3	1,000 objects ILSVRC2012	299x299x3	N/A	1	42.8 ms	Top-1: 79.9% Top-5: 95.7%	No	23.9 MB	Edge TPU model, CPU model, Labels file, All model files
Inception V4	1,000 objects ILSVRC2012	299x299x3	N/A	1	84.7 ms	Top-1: 80.7% Top-5: 95.5%	No	42.9 MB	Edge TPU model, CPU model, Labels file, All model files
MobileNet V1	1,000 objects ILSVRC2012	128x128x3	0.25	1	0.9 ms	Top-1: 40.8% Top-5: 67.2%	Yes	0.7 MB	Edge TPU model, CPU model, Labels file
MobileNet V1	1,000 objects ILSVRC2012	160x160x3	0.5	1	1.4 ms	Top-1: 63.7% Top-5: 83.4%	Yes	1.6 MB	Edge TPU model, CPU model, Labels file
MobileNet V1	1,000 objects ILSVRC2012	192x192x3	0.75	1	1.8 ms	Top-1: 67.2% Top-5: 88.1%	Yes	2.8 MB	Edge TPU model, CPU model, Labels file
MobileNet V1	1,000 objects ILSVRC2012	224x224x3	1.0	1	2.8 ms	Top-1: 69.5% Top-5: 90.6%	Yes	4.4 MB	Edge TPU model, CPU model, Labels file, All model files
MobileNet V2	900+ birds iNaturalist 2017	224x224x3	1.0	1	2.6 ms	N/A	Yes	4.1 MB	Edge TPU model, CPU model, Labels file
MobileNet V2	1000+ insects iNaturalist 2017	224x224x3	1.0	1	2.7 ms	N/A	Yes	4.1 MB	Edge TPU model, CPU model, Labels file
MobileNet V2	2000+ plants iNaturalist 2017	224x224x3	1.0	1	2.6 ms	N/A	Yes	5.5 MB	Edge TPU model, CPU model, Labels file
MobileNet V2	1,000 objects ILSVRC2012	224x224x3	1.0	1	2.9 ms	Top-1: 73.2% Top-5: 90.0%	Yes	4.0 MB	Edge TPU model, CPU model, Labels file, All model files
MobileNet V1	1,000 objects ILSVRC2012	224x224x3	1.0	2	2.8 ms	Top-1: 69.5% Top-5: 89.8%	Yes	4.5 MB	Edge TPU model, CPU model, Labels file
MobileNet V2	1,000 objects ILSVRC2012	224x224x3	1.0	2	3.0 ms	Top-1: 73.2% Top-5: 91.8%	Yes	4.1 MB	Edge TPU model, CPU model, Labels file
MobileNet V3	1,000 objects ILSVRC2012	224x224x3	1.0	2	3.0 ms	Top-1: 77.5% Top-5: 93.6%	Yes	4.9 MB	Edge TPU model, CPU model, Labels file
ResNet-50	1,000 objects ILSVRC2012	224x224x3	N/A	2	42.2 ms	Top-1: 73.6% Top-5: 93.8%	No	25.0 MB	Edge TPU model, CPU model, Labels file
Popular Products V1 New	100,000 popular US products	224x224x3	N/A	1	7.0 ms	N/A	Yes	9.8 MB	Edge TPU model, CPU model, TF Hub model, Labels file

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

² 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.)

^* Beware that the EfficientNet family of models have unique input quantization values (scale and zero-point) that you must use when preprocessing your input. For example preprocessing code, see the classify_image.py or classify_image.cc examples.

On-device training models

These models are designed for compatibility with the on-device transfer learning APIs provided with PyCoral and libcoral.

The "backpropagation" models are embedding extractor models, compiled with the last fully-connected layer removed. They do not perform classification on their own, and must be paired with the SoftmaxRegression API, which allows you to perform on-device backpropagation to train the classification layer.

The "weight imprinting" models are modified to include an L2-normalization layer and other changes to be compatible with the ImprintingEngine API, which performs weight imprinting to retrain classifications.

Model name	Training style	Base dataset	Input size	TF ver.	Micro ¹	Model size	Downloads
EfficientNet-EdgeTpu (L)	Backpropagation	1,000 objects ILSVRC2012	300x300x3	1	Yes	11.7 MB	Edge TPU model, CPU model, Labels file
EfficientNet-EdgeTpu (M)	Backpropagation	1,000 objects ILSVRC2012	240x240x3	1	Yes	7.6 MB	Edge TPU model, CPU model, Labels file
EfficientNet-EdgeTpu (S)	Backpropagation	1,000 objects ILSVRC2012	224x224x3	1	Yes	5.7 MB	Edge TPU model, CPU model, Labels file
MobileNet V1	Backpropagation	1,000 objects ILSVRC2012	224x224x3	1	Yes	3.5 MB	Edge TPU model, CPU model, Labels file
MobileNet V1	Weight imprinting	1,000 objects ILSVRC2012	224x224x3	1	No	5.3 MB	Edge TPU model, CPU model, Labels file, All model files

¹ 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.)

Example code

Basic image classification

An example that performs image classification with a single photo.

Languages: Python, C++

View Python on GitHub View C++ on GitHub

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, C++

View on GitHub

Pipelined image classification

An example showing how to pipeline a model across multiple Edge TPUs, allowing you to significantly increase throughput for large models such as Inception.

Languages: Python, C++

View Python on GitHub View C++ on GitHub

Training scripts

If you’d like to train a classification model to recognize new objects, try the following tutorials:

Trained models link

On-device training models link

Example code link

Training scripts link

More resources link

Trained models

On-device training models

Example code

Training scripts

More resources