ConvESN

Implementation of Convolutional Echo State Network for human activity recognition as outlined in https://www.ijcai.org/proceedings/2017/0342.pdf

This now works with MSRDailyActivity dataset.

Data Setup

mkdir data reservoir check_points && cd data && mkdir DataBackUp MSRDailyAct3D padded modified

The skeleton data files can be obtained here. The Dataset ought to be saved at data/MSRDailyAct3D/

Data Pre-processing

python src/Data_Preparation/daily_activity_skeletons_to_msr_actions.py ./data/MSRDailyAct3D/ ./data/modified/ --folder

this snippet modifies MSRDailyActivityData into a format accepted by the ConvESN code

python src/Data_Preparation/Load_MSRA3D_real_world_modified.py ./data/modified/

add --test and -output to handle a single file

python src/Data_Preparation/Padding.py ./data/DataBackUp ./data/padded/

pads the input data

Train network v2.0

python src/MSMC_2_0.py config/train_config.yaml

Ignore these (testing code snipets)

x="./data/MSRDailyAct3D/a13_s06_e01_skeleton.txt"
python src/Data_Preparation/daily_activity_skeletons_to_msr_actions.py $x ./data/test/

python src/Data_Preparation/plot_dailyactivity3d_skeleton.py $x

python src/MSMC_sai.py ./data/padded/ -split_number 1 -checkpoint check_points/test.hdf5 -reservoir reservoir/rs100.pkl --train

omit --train, to set mode to test add --test_sample, to will work on a single file