problem of train/test list

[shuangshuangguo]

trafficstars

Hi, @hx173149 Thank you for sharing the wonderful work! It seems that you don't follow the official splits, you just make the training/testing ratio 3:1. Am I right? Thanks.

[LiangXu123]

it's really does not matter,since all models are trained and tested on standard UCF101 split1 train/test, so you can prepare your own train/test list as the paper says instead of using the list produced by this repository . as what I write in the README: Note: 1.All report results are done specific on UCF101 split1 (train videos:9537,test videos:3783).

[PBRAOS]

So that i understand.... if the split is done in such a way that the TESTING set has horse riding and the TRAINING set does not have, this does not matter ? You need to train horse riding to the model so that you can benchmark it after.... I think the file that does the file splitting lists has an error.