omarsayed7 / deep-emotion Goto Github PK

Hi, thanks for the contribution
You have the training prepared for FER2013 right? According o the hyperparameters selected in the paper, they are wrong.. for example line 25 in deep_emotion.py. Shouldnt be kernel_size=3 instead of 7? I dont know if there are other mistakes.
Another one, test.py doesnt have emotion column, so I had to use icml_face_data.csv instead to split for train, valid and test.
And last one, I want to know if you can pass me the hyperparameters for each dataset? Because before each modification I got at most 52% test accuracy for FER2013.
Greetings

Why in test.csv dataset of fer2013 do it search for emotion column? Test.csv doesnt have emotion column.. there i show the issue:

File "main.py", line 104, in
validation_dataset= Plain_Dataset(csv_file=validationcsv_file, img_dir = validation_img_dir, datatype = 'val', transform = transformation)
File "/content/Deep-Emotion/data_loaders.py", line 23, in init
self.lables = self.csv_file['emotion']
File "/usr/local/lib/python3.7/dist-packages/pandas/core/frame.py", line 2906, in getitem
indexer = self.columns.get_loc(key)
File "/usr/local/lib/python3.7/dist-packages/pandas/core/indexes/base.py", line 2900, in get_loc
raise KeyError(key) from err
KeyError: 'emotion'

Hello, why is my accuracy rate on fer2013 data set only around 47%? I didn't make any changes to your code

Dear Omar Sayed,
I enjoyed reviewing your implementation of Deep-Emotion.
With your permission, I will appreciate if you share a pretrained model / a .pt file.
Thank you

The spatial transformer network was applied to the input ?
Only 55% acc on FER2013 Database ?
WTF ?
What's wrong with it ???

After I successfully run the command (python main.py -s setup -d data), I got the corresponding photos and table. However, when I run the command (python main.py -t train -d data), I still did not start the training.

The error was as follows : KeyError: 'emotion'
Traceback (most recent call last):
File "main.py", line 104, in
validation_dataset= Plain_Dataset(csv_file=validationcsv_file, img_dir = validation_img_dir, datatype = 'val',
transform = transformation)
self.lables = self.csv_file['emotion']
indexer = self.columns.get_loc(key)
raise KeyError(key) from err
KeyError: 'emotion'

How to solve this problem, thank you

Dear Omer Sayed

Thank you for sharing the implementation.

In the diagram explaining the architecture, it is shown that the output of the grid generation is combined with the output of feature extraction. However, in your implementation, the output of the grid generation is given to the feature extraction layer as input.

Do I miss something?

The spatial transformation network proposed in this paper is applied to the last layer's feature maps, but it is applied to the input image in the code provided, and I used the fer2013 dataset to train the network, but I only got 55% accuracy, which is much lower than the data provided in the paper. Why? I hope you can tell me about it. Thank you very much!

Dear Sir,
Requesting you to share the list of images used for training, validation and test set for Jaffe database. It will help me to compare the results of our algorithm.
Thank you

Just wanted to ask if the pre-trained model ('.pt' or '.pth' files) on any of the datasets are available, and if so where I would be able to find them.

hello, I'm a student who is learning AI.
I was download your code, but I have a problem about this code,
The paper saids "The loss function in this work is simply the summation of two terms, the classification loss(cross-entropy), and the regularization term(which is l2 norm of the weights in the last two fully-connected layers)"
In the main.py, I just see "the classification loss". Is there any particular reason for this?

finally, thank you for your code contributions

Hello,
Thank you for sharing this code,

I was wondering how to have the confusion matrix like those published in the research article please ?

Thank you for your answer

when train the model : python main.py -t True --data "data"
every time got an error : target XXX is out of bounds (The numbers xxx that have appeared are: 3579/541/... )

I have tried to run the code with the train and test data from Kaggle, but it went wrong when running the main.py for training the model, and I found the problem is that the plain_dataset() in data_loaders.py trying to retrieve the ['emotion'] column in val.csv which doesn't exist. How can I solve this problem?

Please I need a pre-trained model, if you have one.

Thanks in advance.

Please I need a pre-trained model...

Thanks in advance.