MobileNet-MultiHW-AVG (MediaPipe)
MobileNet-MultiHW-AVG is a CNN architecture optimized to run well on a variety of hardware while still achieving state of the art performance on Vision ML tasks.MobileNet-MultiHW-AVG image object detection model is a RetinaNet model using a MobileNet-MultiHW-AVG backbone.
The RetinaNet model is a one-stage object detection model which is designed to run faster than typical two-stage models while maintaining the same performance. RetinaNet utilizes a Feature Pyramid Network(FPN) along with a Convolutional Neural Network(CNN) backbone such as MobileNetV2 to create feature maps over the input image at different scales. For more information, see "Focal Loss for Dense Object Detection" by Lin et al (2017).
MobileNet-MultiHW-AVG is a CNN architecture optimized to run well on a variety of hardware while still achieving state of the art performance on Vision ML tasks. Built upon previous models in the MobileNet architecture family such as MobileNetV2, MobileNet-MultiHW-AVG uses neural architecture search(NAS) to optimize for inference latency and accuracy on a variety of hardware including CPU, GPU, DSP, and EdgeTPU. For more information, see "Discovering Multi-Hardware Mobile Models via Architecture Search" by Chu et al (2020).
Using the notebook, you can create a custom MobileNet-MultiHW-AVG model in TFLite, which can be deployed on-device (Android, iOS, Web, desktop, etc) using MediaPipe Tasks ImageClassifier. Use MediaPipe Studio to evaluate the model through interactive live demo.
The demo below is for internal testing purposes only. Output should not be saved or distributed. Please do not provide personally identifiable information or other data subject to regulatory requirements.
This model can be used in a notebook. Click Open notebook to use the model in Colab.
This model checkpoint was pre-trained on the Common Objects in Context(COCO) dataset.
Training images are resized/rescaled to the same resolution (256, 256) and normalized by mean and standard deviation values of 127.5. For more details, see the notebook.
Given an input image, the model will output bounding boxes and object classes with confidence scores.
Resource ID | Release date | Release stage | Description |
---|---|---|---|
mediapipe/MobileNet-MultiHW-AVG-001 | 2024-04-01 | General Availability | Fine tuning and ondevice serving |
O console do Google Cloud não conseguiu carregar fontes JavaScript a partir de www.gstatic.com.
Os possíveis motivos são: