Facial expression recognition in python
Budget: $10 – $30 USD
I want to make a facial expression recognition model using different datasets. I have found a code on Github with 65.5% accuracy (https://github.com/rishabhjainps/Facial-Expression-Recognition). This code works with the fer2013.csv dataset. I want to test the same model with different datasets like:
https://faces.mpdl.mpg.de/imeji/
https://zenodo.org/record/3451524
https://fei.edu.br/~cet/facedatabase.html
https://susanqq.github.io/UTKFace/
https://sites.google.com/andrew.cmu.edu/tarrlab/resources
What you first need to do is an application that will recognize different types of emotions, using different data sets. The accuracy with which emotions are recognized (in different data sets) should be determined and a comparative analysis should be made. It is also important that the data sets contain a similar number of data/images.
The model is already trained, so the first thing you need to do is let the code work with the specified data sets and see how accurately it recognizes your emotions in each data set. Then you need to train the model with another data set (another database, for example, The Japanese Female Facial Expression (JAFFE) Database), and then run it again with the other datasets. The goal is to get the best results, ie once trained a model should give the best possible results - better matches/recognitions, in all data sets.
Fer2018 can be used instead of 2013 if it shows better results. https://www.kaggle.com/datasets?search=FER2018
The code that I posted from Github can be used too.
https://faces.mpdl.mpg.de/imeji/
https://zenodo.org/record/3451524
https://fei.edu.br/~cet/facedatabase.html
https://susanqq.github.io/UTKFace/
https://sites.google.com/andrew.cmu.edu/tarrlab/resources
What you first need to do is an application that will recognize different types of emotions, using different data sets. The accuracy with which emotions are recognized (in different data sets) should be determined and a comparative analysis should be made. It is also important that the data sets contain a similar number of data/images.
The model is already trained, so the first thing you need to do is let the code work with the specified data sets and see how accurately it recognizes your emotions in each data set. Then you need to train the model with another data set (another database, for example, The Japanese Female Facial Expression (JAFFE) Database), and then run it again with the other datasets. The goal is to get the best results, ie once trained a model should give the best possible results - better matches/recognitions, in all data sets.
Fer2018 can be used instead of 2013 if it shows better results. https://www.kaggle.com/datasets?search=FER2018
The code that I posted from Github can be used too.