nku-zhichengzhang / cten Goto Github PK

Hello,
I am thoroughly reviewing the details mentioned in your paper, "Weakly Supervised Video Emotion Detection and Prediction via Cross-Modal Temporal Erasing Network," and attempting to replicate your experimental results from your open-source GitHub repository. In the paper, you mentioned: "We train the parameter of the attention module and linear classifier by a learning rate of 2e-4."

I have noticed that you set the default epoch to 100 in the code, but I have some questions regarding the adjustment of the learning rate during the training process and the number of training epochs to achieve the final results on the VideoEmotion-8 dataset. Specifically:

1. Did the learning rate of 2e-4 remain constant throughout the entire 100 epochs of training, as mentioned in the paper?
2. Were there any adjustments made to the learning rate during the experiment? If so, how were these adjustments made?
3. Could you provide details on the learning rate and the number of training epochs used to obtain the final experimental results on the VideoEmotion-8 dataset?

I did not find specific information on these details in the paper, and I hope to gain insights into these aspects to better understand your work and achieve similar results in my experiments.

Thank you!

Thanks for your amazing repo
I want to run your code, but I don't know how to get the "3d resnet101" checkpoint.
Could you please provide the download link?

def generate_visual_Erase_model(opt):
model=VisualErase(
snippet_duration=opt.snippet_duration,
sample_size=opt.sample_size,
n_classes=opt.n_classes,
seq_len=opt.seq_len,
pretrained_resnet101_path=opt.resnet101_pretrained,
)
model = nn.DataParallel(model)
model=model.cuda()
return model, model.parameters()

Hello, I have a question about this model, you input is video frames and audio in your papers, but this model input just video frames. So I need your help of this question. If this is not the right model can you tell me the right one. Thanks.

Hi! I'm very interested in your great work. I have some questions as below:

1. I have a problem like here. How to deal with these mp3 files?
2. Can you also provide a dataloader for ek6 and CAER datasets?
3. In contrast to VAANet, the current code in this git doesn't utilize audio input (model code). Where can I find complete code that uses both inputs, visual and audio?
   Thx

hello! I'm very interested in your work. Could you please share me the pretrained resnet101 model? Thank you.