VolvoConstScene_HAR_SFU

Repository for the paper "Language Supervised Human Action Recognition with Salient Fusion: Construction Worker Action Recognition as a Use-Case"

Prerequisites

• Python >= 3.6

• PyTorch >= 1.1.0

• PyYAML, tqdm, tensorboardX

• We provide the dependency file of our experimental environment, you can install all dependencies by creating a new anaconda virtual environment and running pip install -r requirements.txt and pip install -r requirements2.txt

pip install torch==1.12.1+cu116 torchvision==0.13.1+cu116 torchaudio==0.12.1 --extra-index-url https://download.pytorch.org/whl/cu116

Data Preparation

You can download NTU and NTU120 datasets from NTU and follow CTR-GCN for data preparation. Also, NTU and NTU120 cropped images are provided here:

NTU and NTU120

VolvoConstAct dataset link: Dataset

Train

For NTU60:

python main_train_ntu.py

For NTU120:

python main_train\ntu120.py

For ConsAct:

python main_train_volvo.py

Load Pretrained

python main_train_volvo.py --load_pretrained True --pretrained_address path/to/address