Code for the paper Action Sets: Weakly Supervised Action Segmentation without Ordering Constraints

• download the data from https://uni-bonn.sciebo.de/s/wOxTiWe5kfeY4Vd
• extract it so that you have the data folder in the same directory as main.py
• create a results directory in the same directory where you also find main.py: mkdir results

Requirements: Python3.x with the libraries numpy, pytorch (version 0.4.1), and scipy

Run python main.py training

Run python main.py inference --n_threads=NUM_THREADS, where NUM_THREADS should be replaced with the number of parallel CPU threads you want to use for Viterbi decoding.

In the inference step, recognition files are written to the results directory. The frame-level ground truth is available in data/groundTruth. Run python eval.py --recog_dir=results --ground_truth_dir=data/groundTruth to evaluate the frame accuracy of the trained model

We provide a python/pytorch implementation for easy usage. In the paper, we used a faster, in-house C++ implementation, so results can be slightly different. Running the provided setup on split1 of Breakfast should lead to roughly 23% frame accuracy.

If you use the code, please cite

A. Richard, H. Kuehne, J. Gall:
Action Sets: Weakly Supervised Action Segmentation without Ordering Constraints
in IEEE Int. Conf. on Computer Vision and Pattern Recognition, 2018