Adamink/ViewAdaptiveLSTM

This project reimplements View-adaptive neural network and transfers it to PKUMMD dataset.

This project implements View-Adaptive Neural Network on PKUMMD dataset.

The codes involve implementation of the following paper:

• View Adaptive Recurrent Neural Networks for High Performance Human Action Recognition from Skeleton Data

Dataset:

Change paths in scripts/global_variable.py to path to normalized PKUMMD-v1 data, and path to raw PKUMMD-v2 data and label. then run

cd scripts
python gen_v1.py
python gen_v1_strans.py
python gen_v2.py
python gen_v2_strans.py

Run

Change --dataset_dir in scripts/v?_?.sh to path to corresponding destination in scripts/global_variable.py, then run

cd $PATH_TO_FOLDER$
sh ./scripts/v1_CS.sh
sh ./scripts/v1_L.sh
sh ./scripts/v1_M.sh
sh ./scripts/v1_R.sh
sh ./scripts/v2_CS.sh
sh ./scripts/v2_L.sh
sh ./scripts/v2_M.sh
sh ./scripts/v2_R.sh

Results

Part1 norm 
L 85.4
R 84.1
M 93.2
CS 86.7

Part2 no norm
MR/L 33.3
LR/M 40.3
LM/R 29.7
CS 49.0