Cathy
Cathy
I have the same question. How did you gain such significant improvement from the original implementation? Is it because you use multiple GPUs to train? Thank you very much!
Hi, @Michael-Ray-Sirius! Please check out [this link](https://drive.google.com/drive/folders/1S3Ub0snH2-71OgGViIGdgHYqdRmsKrYV?usp=sharing) for the I3D features used in the project. The features are being uploaded and they should be ready in 1 hour.
I've updated the Anet code in the repo. Please just refresh and check out the branch: [VSGN_ActivityNet](https://github.com/coolbay/VSGN/tree/VSGN_ActivityNet)