The MediaPipe Image Embedder task lets you create a numeric representation of an image,
which is useful in accomplishing various ML-based image tasks. This
functionality is frequently used to compare the similarity of two images using
mathematical comparison techniques such as Cosine Similarity. This task operates
on image data with a machine learning (ML) model as static data or a continuous
stream, and outputs a numeric representation of the image data as a list of
high-dimensional feature vectors, also known as embedding vectors, in either
floating-point or quantized form.
Start using this task by following one of these implementation guides for your
target platform. These platform-specific guides walk you through a basic
implementation of this task, using a recommended model, and provide code
examples with the recommended configuration options:
This section describes the capabilities, inputs, outputs, and configuration
options of this task.
Features
Input image processing - Processing includes image rotation, resizing,
normalization, and color space conversion.
Region of interest - Performs embedding on a region of the image instead
of the whole image.
Embedding similarity computation - Built-in utility function to compute
the cosine similarity
between two feature vectors
Quantization - Supports scalar quantization for the feature vectors.
Task inputs
Task outputs
Input can be one of the following data types:
Still images
Decoded video frames
Live video feed
Image Embedder outputs a list of embeddings consisting of:
Embedding: the feature vector itself, either in floating-point form or
scalar-quantized.
Head index: the index for the head that produced this embedding.
Head name (optional): the name of the head that produced this embedding.
Configurations options
This task has the following configuration options:
Option Name
Description
Value Range
Default Value
running_mode
Sets the running mode for the task. There are three
modes:
IMAGE: The mode for single image inputs.
VIDEO: The mode for decoded frames of a video.
LIVE_STREAM: The mode for a livestream of input
data, such as from a camera. In this mode, resultListener must be
called to set up a listener to receive results
asynchronously.
{IMAGE, VIDEO, LIVE_STREAM}
IMAGE
l2_normalize
Whether to normalize the returned feature vector with L2 norm.
Use this option only if the model does not already contain a native
L2_NORMALIZATION TFLite Op. In most cases, this is already the case and
L2 normalization is thus achieved through TFLite inference with no need
for this option.
Boolean
False
quantize
Whether the returned embedding should be quantized to bytes via
scalar quantization. Embeddings are implicitly assumed to be unit-norm and
therefore any dimension is guaranteed to have a value in [-1.0, 1.0]. Use
the l2_normalize option if this is not the case.
Boolean
False
result_callback
Sets the result listener to receive the embedding results
asynchronously when the Image Embedder is in the live stream
mode. Can only be used when running mode is set to LIVE_STREAM
N/A
Not set
Models
The Image Embedder requires an image embedding model to be downloaded and stored
in your project directory. Start with the default, recommended model for your
target platform when you start developing with this task. The other available
models typically make trade-offs between performance, accuracy, resolution, and
resource requirements, and in some cases, include additional features.
MobileNetV3 model
This model family uses a MobileNet V3
architecture and was trained using ImageNet data.
This model uses a multiplier of 0.75 for the depth (number of features) in the
convolutional layers to tune the accuracy-latency trade off. In addition,
MobileNet V3 comes in two different sizes, small and large, to adapt the network
to low or high resource use cases.
Here's the task benchmarks for the whole pipeline based on the above
pre-trained models. The latency result is the average latency on Pixel 6 using
CPU / GPU.
[[["Easy to understand","easyToUnderstand","thumb-up"],["Solved my problem","solvedMyProblem","thumb-up"],["Other","otherUp","thumb-up"]],[["Missing the information I need","missingTheInformationINeed","thumb-down"],["Too complicated / too many steps","tooComplicatedTooManySteps","thumb-down"],["Out of date","outOfDate","thumb-down"],["Samples / code issue","samplesCodeIssue","thumb-down"],["Other","otherDown","thumb-down"]],["Last updated 2025-01-13 UTC."],[],[],null,["# Image embedding task guide\n\nThe MediaPipe Image Embedder task lets you create a numeric representation of an image,\nwhich is useful in accomplishing various ML-based image tasks. This\nfunctionality is frequently used to compare the similarity of two images using\nmathematical comparison techniques such as Cosine Similarity. This task operates\non image data with a machine learning (ML) model as static data or a continuous\nstream, and outputs a numeric representation of the image data as a list of\nhigh-dimensional feature vectors, also known as embedding vectors, in either\nfloating-point or quantized form.\n\n[Try it!arrow_forward](https://mediapipe-studio.webapps.google.com/demo/image_embedder)\n\nGet Started\n-----------\n\nStart using this task by following one of these implementation guides for your\ntarget platform. These platform-specific guides walk you through a basic\nimplementation of this task, using a recommended model, and provide code\nexamples with the recommended configuration options:\n\n- **Android** - [Code\n example](https://github.com/google-ai-edge/mediapipe-samples/tree/main/examples/image_embedder/android) - [Guide](./android)\n- **Python** - [Code\n example](https://colab.sandbox.google.com/github/googlesamples/mediapipe/blob/main/examples/image_embedder/python/image_embedder.ipynb) - [Guide](./python)\n- **Web** - [Code example](https://codepen.io/mediapipe-preview/pen/yLqxbaZ) - [Guide](./web_js)\n\nTask details\n------------\n\nThis section describes the capabilities, inputs, outputs, and configuration\noptions of this task.\n\n### Features\n\n- **Input image processing** - Processing includes image rotation, resizing, normalization, and color space conversion.\n- **Region of interest** - Performs embedding on a region of the image instead of the whole image.\n- **Embedding similarity computation** - Built-in utility function to compute the [cosine similarity](https://en.wikipedia.org/wiki/Cosine_similarity) between two feature vectors\n- **Quantization** - Supports scalar quantization for the feature vectors.\n\n| Task inputs | Task outputs |\n|-------------------------------------------------------------------------------------------------------------------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|\n| Input can be one of the following data types: - Still images \u003c!-- --\u003e - Decoded video frames \u003c!-- --\u003e - Live video feed | Image Embedder outputs a list of embeddings consisting of: - Embedding: the feature vector itself, either in floating-point form or scalar-quantized. \u003c!-- --\u003e - Head index: the index for the head that produced this embedding. \u003c!-- --\u003e - Head name (optional): the name of the head that produced this embedding. |\n\n### Configurations options\n\nThis task has the following configuration options:\n\n| Option Name | Description | Value Range | Default Value |\n|-------------------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|-------------------------------|---------------|\n| `running_mode` | Sets the running mode for the task. There are three modes: \u003cbr /\u003e IMAGE: The mode for single image inputs. \u003cbr /\u003e VIDEO: The mode for decoded frames of a video. \u003cbr /\u003e LIVE_STREAM: The mode for a livestream of input data, such as from a camera. In this mode, resultListener must be called to set up a listener to receive results asynchronously. | {`IMAGE, VIDEO, LIVE_STREAM`} | `IMAGE` |\n| `l2_normalize` | Whether to normalize the returned feature vector with L2 norm. Use this option only if the model does not already contain a native L2_NORMALIZATION TFLite Op. In most cases, this is already the case and L2 normalization is thus achieved through TFLite inference with no need for this option. | `Boolean` | `False` |\n| `quantize` | Whether the returned embedding should be quantized to bytes via scalar quantization. Embeddings are implicitly assumed to be unit-norm and therefore any dimension is guaranteed to have a value in \\[-1.0, 1.0\\]. Use the l2_normalize option if this is not the case. | `Boolean` | `False` |\n| `result_callback` | Sets the result listener to receive the embedding results asynchronously when the Image Embedder is in the live stream mode. Can only be used when running mode is set to `LIVE_STREAM` | N/A | Not set |\n\n| **Note:** The `resultListener` is dependent on `runningMode`. Only set the `resultListener` when `runningMode` is set to `LIVE_STREAM`.\n\nModels\n------\n\nThe Image Embedder requires an image embedding model to be downloaded and stored\nin your project directory. Start with the default, recommended model for your\ntarget platform when you start developing with this task. The other available\nmodels typically make trade-offs between performance, accuracy, resolution, and\nresource requirements, and in some cases, include additional features.\n| **Attention:** This MediaPipe Solutions Preview is an early release. [Learn more](/edge/mediapipe/solutions/about#notice).\n\n### MobileNetV3 model\n\nThis model family uses a [MobileNet V3](https://arxiv.org/abs/1905.02244)\narchitecture and was trained using [ImageNet](https://www.image-net.org) data.\nThis model uses a multiplier of 0.75 for the depth (number of features) in the\nconvolutional layers to tune the accuracy-latency trade off. In addition,\nMobileNet V3 comes in two different sizes, small and large, to adapt the network\nto low or high resource use cases.\n\n| Model name | Input shape | Quantization type | Versions |\n|----------------------------------------------------------------------------------------------------------------------------------------------------|-------------|-------------------|--------------------------------------------------------------------------------------------------------------------------------------|\n| [MobileNet-V3 (small)](https://storage.googleapis.com/mediapipe-models/image_embedder/mobilenet_v3_small/float32/latest/mobilenet_v3_small.tflite) | 224 x 224 | None (float32) | [Latest](https://storage.googleapis.com/mediapipe-models/image_embedder/mobilenet_v3_small/float32/latest/mobilenet_v3_small.tflite) |\n| [MobileNet-V3 (large)](https://storage.googleapis.com/mediapipe-models/image_embedder/mobilenet_v3_large/float32/latest/mobilenet_v3_large.tflite) | 224 x 224 | None (float32) | [Latest](https://storage.googleapis.com/mediapipe-models/image_embedder/mobilenet_v3_large/float32/latest/mobilenet_v3_large.tflite) |\n\nTask benchmarks\n---------------\n\nHere's the task benchmarks for the whole pipeline based on the above\npre-trained models. The latency result is the average latency on Pixel 6 using\nCPU / GPU.\n\n| Model Name | CPU Latency | GPU Latency |\n|----------------------|-------------|-------------|\n| MobileNet-V3 (small) | 3.94ms | 7.83ms |\n| MobileNet-V3 (large) | 9.75ms | 9.08ms |"]]
