search

wenliangdai / Multimodal-End2end-Sparse

The code repository for NAACL 2021 paper "Multimodal End-to-End Sparse Model for Emotion Recognition".
95 stars 16 forks source link

can't reproduce the paper's performance #18

Open JYeonKim opened 1 year ago

JYeonKim commented 1 year ago

Hi, Thank you very much for the code you shared.

I tried to run the code with the following arguments:

python main.py -lr=5e-5 -ep=40 -mod=tav -bs=8 --img-interval=500 --early-stop=6 --loss=bce --cuda=3 --model=mme2e_sparse --num-emotions=6 --trans-dim=64 --trans-nlayers=4 --trans-nheads=4 --text-lr-factor=10 -st=0.7 --text-model-size=base --text-max-len=100 --dataset="iemocap"

The papers syas that we get an accuracy of 84.4% but the best accuracy that I am getting is 83.0% on the IEMOCAP dataset. Could you please put the exact arguments to reproduce the results of the papers? How can I get the results of the papers?

(I think "p" equals "st" argument. So I set "st=0.7")

ACC ang exc fru hap neu sad average
paper(p=0.7) 88.2 88.3 74.9 89.5 77.0 88.6 84.4
me(p=0.7) 88.8 80.6 77.9 90.4 74.5 86.0 83.0
me(p=0.9) 88.3 87.5 76.7 90.2 72.1 90.6 84.2

best regards, Juyeon Kim

rafiepour commented 1 year ago

Hi, same thing is happening for me. Any updates?