search

facebookresearch / supervision-by-registration

Supervision-by-Registration: An Unsupervised Approach to Improve the Precision of Facial Landmark Detectors
Other
762 stars 165 forks source link

Train my own dataset #55

Closed abdou31 closed 4 years ago

abdou31 commented 5 years ago

Issue description

Can I use this project (supervision-by-registration) to train my own dataset? If yes, what instructions should I follow? Thanks

System Info

Windows 10 x64

D-X-Y commented 5 years ago

You need to generate a file list of your dataset, including some necessary information such as image-path, face bounding box, landmark locations. An example of 300-W is at https://github.com/facebookresearch/supervision-by-registration/blob/master/cache_data/generate_300W.py.

abdou31 commented 5 years ago

Actually , the result that I hope to get looks like this:

dsaa

Is it possible with this tool? If yes, that's good but maybe the one thing that can be so hard is labelling image according to this result ( labelling the iris region )

D-X-Y commented 5 years ago

Yes, this repo can train a detector to predict eye-landmarks. But you need to feed it with the corresponding annotations for training.

abdou31 commented 5 years ago

I have two questions:

D-X-Y commented 5 years ago

1, What device do you plan to run the model on for real-time-detection?

2, The model can be used to detect, while this repo does not support the Andrioid platform.

abdou31 commented 5 years ago
  1. I want to run the model for real-time-detection on Android device (e.g Samsung A3 2016).
  2. Is it supported in the Android platform? ( I mean the extension of the model, e.g tflite* extension is supported on Android device )
    • *tflite extension is the extension of Tensorflow model on the device.
AaronXinWu commented 5 years ago
  1. I want to run the model for real-time-detection on Android device (e.g Samsung A3 2016).
  2. Is it supported in the Android platform? ( I mean the extension of the model, e.g tflite* extension is supported on Android device )
  • *tflite extension is the extension of Tensorflow model on the device. same question
D-X-Y commented 5 years ago

Sorry for the late reply. The model might be possible to run on Android device, but we did not test it on edge devices. You need to convert the model to tflite models so as to run on Android platform.