Video sub-action parsing algorithm (ASP) on the frisby-catch action of THUMOS2014 dataset.

1.Please unzip the file codeUploadNew.rar. The code, data and result dirs represent the Matlab version of algorithm (Matlab 2014a is adopted), all the features and groundtruth of dataset and the running result of our ASP method.

2.In dir 'code', a_main.m is the main function. Before you start the a_main.m function. You should place your feature of dataset under the dir 'stipFeatureThumos'. An example of the feature file is included in the `stipFeatureThumos'.

3. In dir 'data', dir 'parameters' includes some parameters of ASP and SP. 'videoGrounTruth' is the ground truth of THUMOS'14: ambiguous segments, framerate and frames of videos, and the ground truth segments.'codeAll' and `confAll' are the BoW coding we've learned.

4.A video showing the result of ASP is attached together with its explanation doc file.

5.The paper 'Soft Video Parsing by Label Distribution Learning' is published at AAAI17. The extended version would be published at Frontiers of Computer Science. If you have any question, please contact [e-mail] or [e-mail].