search

HYPJUDY / Decouple-SSAD

Decoupling Localization and Classification in Single Shot Temporal Action Detection
https://arxiv.org/abs/1904.07442
MIT License
96 stars 19 forks source link

Enter a single unlabeled video to detect the input point #26

Open mrlihellohorld opened 3 years ago

mrlihellohorld commented 3 years ago

Hi, thank you and your team for your contribution to this project. I have trained a D-SSAD model based on my own data set according to your code, but how can I infer a video without tags and cropped according to the trained model?

HYPJUDY commented 3 years ago

Hi, I think you could treat the process as testing stage. You can pre-process the video and predict labels for it with your trained model with the testing function.

mrlihellohorld commented 3 years ago

Hi, I think you could treat the process as testing stage. You can pre-process the video and predict labels for it with your trained model with the testing function.

Hi, Thanks for u reply. I went through the process according to your suggestion, but the prediction I got was poor. May I take the liberty to e-mail you a copy of the results of my training and prediction? could u help me to analyze which step went wrong. With best wishes.