floretcat / cmran Goto Github PK

Hi,
Thank you for the wonderful work.

You extracted features from Resnet in your experiments, where can I find that features?

I would appreciate it if you could answer.

hi~thanks for your work first!
The features in this repo are vgg19 features. Are there any features of resnet151？
If I extract the resnet151 features by myself, what are the video of noisy features?

您好，我在看您论文时发现您用于resnet来提取特征，可是我在处理数据的时候发现只有4097个，不太清楚visual_feature.h5中怎么有4143个视觉数据（数据集备注信息中提到有些视频包含不同的音视频事件所以不是4143个视频），您能告诉我是怎么处理的吗？希望您回复，谢谢！

I cannot use the resnet151 features I extracted myself to reproduce the results in the paper

Hi!
Thanks a lot for sharing the code generously, especially when there are few available codes in this task! I believe your codes will help many researchers in this field, including me.
However, I have a small question about your evaluation method. It seems that the validation set is not used in the training process. Instead, the test set is directly used during every training epoch and the best performance in the test set is chosen as the reported performance. I wonder if this training strategy could lead to the overfitting on the test set, and what's the difference between the test phase and the test progress in the training phase by doing so?
Please remind me if I misunderstand the training process in the released codes, and thanks again for your generous sharing!