如何对单个视频进行预测 about cvpr2022-ftcl HOT 1 CLOSED

First of all, thank you for your interest in our work. If the single video you are talking about is an untrimmed raw video without any pre-processing, you need to convert it into 2048-dimensional features with a pre-trained I3D model. For the pre-trained feature extraction, we refer you to the official TensorFlow-based I3D repo and the third-party PyTorch-based I3D repo. There are detailed operations guides for the feature extraction. If the single video talked about is just one of the testing videos of the THUMOS14 dataset, simply modifying the dataset section of the code will work.

from cvpr2022-ftcl.