maxbbraun / thermal-face Goto Github PK

Hi,

I'm trying to perform inference using the uncompiled model thermal_face_automl_edge_fast.tflite and the TF Lite API as linked, as I'm trying to run the code on my Windows computer. The output from running interpreter.get_tensor(...) appears to be a 1 by 500 by 4 array of floats (in addition, interpreter.get_output_details()[0]['shape'] returns [1 500 4]). I'm not sure how to convert this (1, 500, 4) array into bounding box(es) for face detection?

Hello!!

Would like to play with this, and im wondering if is possible to provide a pb version of the model?

I read that is planned to add landmarks? that would be very cool, do you think is possible?

Regards

I'm new to python, so I'm sorry if it's a stupid question.
I don't have device "EdgeTPU" ...
But I have another thermal camera and a thermal image taken with that camera.
Is face detection possible even in such cases?

Greetings,

I have downloaded the uncompiled model and tried to run the uncompiled model with the script in this link,
but because of some reason I am getting an strange output.

There are 4 outputs with the shapes: ([1, 500,4], [1, 500], [1, 500], [1, 500]) and I don't understand why since in the picture there is just one person.

These are the 2 pictures I tried the model on. If you can provide a testing script for tf-lite I would be grateful or you can just explain how the output shapes are constructed.

Thank you very much for your time.

Something isn't right with the compiled version of the model.

With thermal_face_automl_edge_l.tflite:

With thermal_face_automl_edge_l_edgetpu.tflite:

The FLIR ADAS Thermal Dataset has >10k thermal images, some of which include people. It comes with annotations, but those are bounding boxes around the whole body, not just the face.

It could be useful to get the subset of images with people annotated (e.g. like this) and mix it in with the other training data.

Hello!

I am a new beginner of machine learning. I think this project is very interesting and useful and I would like to perform transfer learning on the tflite models so that I can distinguish different types of thermal faces. May I know if it is possible to train the tflite model on my custom datasets? Or is there any ways to examine the structures and weights of the tflite models so that I can perform transfer learning on them?

Many thanks for your help!

The current approach is to train a detector model from scratch on a dataset that combines visible-light and thermal images, then compile it for Edge TPU.

An alternative approach would be to retrain an existing and already TPU-optimized detector model such as MobileNet SSD v2 (Faces), probably using only the thermal images.

This could be particularly interesting as a workaround for issue #2.

Hi, Maxbbraun

I am very interested in this project and I notice you release the training and compliling code.

If possible, may I know the face detection accuracy on the thermal faces ?

best
Xing

I'd like to use this model with Tensorflow.js. The tflite model is not compatible with tensorflowjs. Could you export a tensorflowjs model?

(yes, I know there is lots of info on training the model, but if you still have it setup in Cloud ML, I thought a polite ask to click the export button might save me the effort of setting up my own Cloud ML instance)

https://medium.com/tensorflow/build-and-deploy-tensorflow-js-models-with-the-power-of-automl-1985b8985083

Hi,

First of all congrats for the awesome project you created. Very appropriate for the current times... :D I am doing some experiments with this Flir camera and a Raspberry and your project works great! The model is able to find my face in the image.

Now I am planning to do some modifications and would like to have a .pb version of the model, instead of the .tflite one. Do you have this .pb version? If so I would be grateful if you could send it to me

Congrats again for the great project.
Cheers!