Minimal code for running inference on spiking neural network trained for Event-based Video Reconstruction via Potential-assisted Spiking Neural Network, CVPR2022.

=======================================================================

Requirements

• Python >= 3.7 (3.9 recommended)
• PyTorch >= 1.6 (1.9 recommended)
• Spikingjelly = 0.0.0.0.6 ======================================================================

Running with Anaconda

cuda_version=10.2

conda create -n snnrec python=3.9 
conda activate snnrec 

conda install -y pytorch torchvision cudatoolkit=$cuda_version -c pytorch
conda install torch torchvision cudatoolkit=$cuda_version

conda install pandas

Install Spikingjelly

pip install spikingjelly==0.0.0.0.6

=====================================================================

Inference

Usage:

conda activate snnrec

python rec_snn.py [-network NETWORK] [-path_to_pretrain_models PATH_TO_PRETRAIN_MODELS] [-path_to_event_files PATH_TO_EVENT_FILES] [-save_path SAVE_PATH] [-height HEIGHT] [-width WIDTH] [-num_events_per_pixel NUM_EVENTS_PER_PIXEL]

# For example, to run EVSNN:
python rec_snn.py -network EVSNN_LIF_final -path_to_pretrain_models ./pretrained_models/EVSNN.pth

# To run PA-EVSNN
python rec_snn.py -network PAEVSNN_LIF_AMPLIF_final -path_to_pretrain_models ./pretrained_models/PAEVSNN.pth

======================================================================

Folder Structure

minimal_code_snn/ | ├── rec_snn.py - evaluation of trained model | ├── data/ - default directory for storing input data | ├── model/ - models, losses, and metrics | ├── dataset.py | ├── snn_network.py | ├── neurons/
| ├── spiking_neuron.py - spiking neurons, MP neurons | ├── results/ - generated results are saved here |
└── utils/ - small utility functions ├── util.py └── ...