yjxiong / anet2016-cuhk

Action Recognition Toolbox for CUHK&ETHZ&SIAT submission to ActivityNet 2016
BSD 2-Clause "Simplified" License
252 stars 92 forks source link

AudioModel_And_SampleMethod #8

Open 3DMM-ICME2023 opened 7 years ago

3DMM-ICME2023 commented 7 years ago

Thanks for your sharing! Congratulations on the 1st in the ActivityNet competition! Could you please answer the following 2 questions.I did not find answers according to your notepaper(http://wanglimin.github.io/contests/XiongW_Anet16.pdf):

  1. The model of rgb and optical flow is resnet and inceptionv3. I want to know what model you used for the audio cnn? alexnet?Train the audio cnn from scratch or finetune the model which pretrained on the imagenet
  2. What is your sample rate when you construct your train dataset: for rgb cnn, 1fps? What about the optical flow cnn?