zerohd4869 / mm-dfn Goto Github PK

Greetings,
i try to reproduce the result on MELD but it doesn't work as i wish (i use the shell script)
and i want to get a detailed parameter settings like --use_topic,--nodal_attention .etc (whether the settings is the same as training on IEMOCAP or not)
thanks!

Thank you for your work !!

I want to train my dataset, what data format is included in the iemocap_features.pkl file of the dataset ?
Is there a sample script file for making iemocap_features.pkl or MELD_features_raw1.pkl ?

Thank you~

I cannot download torch==1.4.0, causing torch_sparse error?

good work. wait for the code.

Dear, this work is nice! Wait for Open source code provided for other workers to learning!

Greetings!
I run the model on IEMOCAP dataset ,however i can not get the result as the paper gives
some parameters are as follows(basically i dont change them)
Namespace(no_cuda=False, dataset='IEMOCAP', data_dir='../data/iemocap/IEMOCAP_features.pkl', multi_modal=True, modals='avl', mm_fusion_mthd='concat_subsequently', use_modal=True, base_model='LSTM', graph_model=True, graph_type='GDF', graph_construct='direct', use_gcn=False, nodal_attention=True, use_topic=True, use_residue=True, av_using_lstm=False, active_listener=False, attention='general', use_crn_speaker=True, speaker_weights='3-0-1', use_speaker=False, reason_flag=False, epochs=30, batch_size=32, valid_rate=0.0, modal_weight=1.0, Deep_GCN_nlayers=16, lr=0.0003, l2=0.0001, rec_dropout=0.1, dropout=0.4, alpha=0.2, lamda=0.5, gamma=0.5, windowp=10, windowf=10, multiheads=6, loss='FocalLoss', class_weight=False, save_model_dir='../outputs/iemocap_demo/', tensorboard=False, test_label=False, load_model='../outputs/iemocap_demo/model_4.pkl', seed=2021)

i wonder what is the detailed parameter to get the aforementioned result
thanks!!!!

Greetings!
i try to train with MELD dataset and change the parameter in :
parser.add_argument('--dataset', default='IEMOCAP', help='dataset to train and test')

and change the IEMOCAP to MELD ,while it reports:
ValueError: not enough values to unpack (expected 10, got 9)

what else can i change
Thanks!!!!

I am currently examining the processed dataset you have shared in your code. In MELD_features_raw1.pkl, I have noted that the second section pertains to speaker information. Each identifier corresponds to a speech segment, and for each segment, there are as many vectors as there are sentences. However, I observed that each vector has a dimension of 9. Could you kindly clarify if the value 9 holds any particular significance? Thank you very much for your assistance.

I found that the MELD dataset you use is 5-classified, and there is no comparison of how you process the two types of data in feature extraction, or how you use to process the two types of data.

The accuracy of 'fear' and 'disgust' are always zero in Meld datasets, so I can't reproduce the result in paper.
The results:

Is there something wrong with my understanding?

Greetings!
i saw the date presented in TABLE.1 in your paper

and i wonder how to get the F1score per class ,Acc and w-F1
are they the data in a best epoch or the result after average since i have noticed that there is a part of best eopch data in your code

Thanks!!!