Tfjs: is there any way to get the layered architecture of the facemesh graph model?

Created on 29 May 2020  Â·  4Comments  Â·  Source: tensorflow/tfjs

To get help from the community, we encourage using Stack Overflow and the tensorflow.js tag.

TensorFlow.js version

Browser version

Describe the problem or feature request

Code to reproduce the bug / link to feature request

If you would like to get help from the community, we encourage using Stack Overflow and the tensorflow.js tag.

GitHub issues for this repository are tracked in the tfjs union repository.

Please file your issue there, following the guidance in that issue template.

facemesh support

All 4 comments

Facemesh model is in graph format https://tfhub.dev/mediapipe/tfjs-model/facemesh/1/default/1
I'd like to see its model architecture. There's nothing much I could find even in the published paper. In the model card https://drive.google.com/file/d/1VFC_wIpw4O7xBOiTgUldl79d9LA-LsnA/view it says "​Convolutional Neural Network: MobileNetV2-like with customized blocks for real-time performance.". Authors have revealed neither the customizations they've done nor the model architecture. Any resources on this would be much appreciated.

Hi @ThejanW - here is a link to a JSON object describing the model's topology: https://storage.googleapis.com/tfhub-tfjs-modules/mediapipe/tfjs-model/facemesh/1/default/1/model.json

Hopefully this helps! These objects are available for all our graph models - if you open the network tab you'll typically see a "model.json" link.

@annxingyuan I humbly appreciate your prompt response. I played around the provided graph model and even converted that to TesorFlow python SavedModel format. I've already looked into the provided model.json. It's apparent that this model topology file is generated using tfjs converter from a TensorFlow SavedModel.

Will you be exposing the TensorFlow model definition scripts or even a tfjs layers model in the near future. From a layers model, what I meant was loading a model.json file with a command like this const model = await tf.loadLayersModel(‘path/to/model.json’);

My goal is to customize this architecture for a different research. But I have to first see the model architecture as layers rather than as ops and nodes to perform my customizations.

Thanks.

Hi @ThejanW - unfortunately, we do not have plans to release a layers version of the model. Here is a link to the TFLite version of the model: https://github.com/google/mediapipe/blob/master/mediapipe/models/face_landmark.tflite

Was this page helpful?
0 / 5 - 0 ratings

Related issues

beele picture beele  Â·  3Comments

KienPM picture KienPM  Â·  3Comments

dhrumil83 picture dhrumil83  Â·  3Comments

chrisdonahue picture chrisdonahue  Â·  3Comments

kylemcdonald picture kylemcdonald  Â·  3Comments