How intelligent-computing-lab-panda/seenn works
SEENN trains spiking neural networks and documents SEENN-I early-exit evaluation; the tree also contains a separate QCFS ANN-to-SNN training/evaluation program and a substantial SpikingJelly runtime. The main workflows are represented alongside the conversion, event-data, neuron, and visualization capabilities visible in the tree. README-documented SEENN-I evaluation has no matching file in the supplied tree, so its implementation path and wiring are not asserted; several root modules are likewise unsampled.
Main entry points: main.py Train and evaluate · train_snn.py SNN training · ann2snn ANN-to-SNN conversion · neuron.py Spiking neurons
Components 19
SEENN training
- SNN training
train_snn.py
QCFS conversion
- Train and evaluate
SNN_conversion_QCFS/main.py - Data preprocessing
SNN_conversion_QCFS/Preprocess - ANN architectures
SNN_conversion_QCFS/Models - ImageNet training
SNN_conversion_QCFS/ImageNet/train.py - ImageNet evaluation
SNN_conversion_QCFS/ImageNet/evaluate.py - Training and evaluation
SNN_conversion_QCFS/funcs.py - Activation conversion
SNN_conversion_QCFS/utils.py
SpikingJelly runtime
- ANN-to-SNN conversion
SNN_conversion_QCFS/spikingjelly/clock_driven/ann2snn - Spiking neurons
SNN_conversion_QCFS/spikingjelly/clock_driven/neuron.py - Neuron kernels
SNN_conversion_QCFS/spikingjelly/clock_driven/neuron_kernel.py - Spike encoding
SNN_conversion_QCFS/spikingjelly/clock_driven/encoding.py - Neuromorphic datasets
SNN_conversion_QCFS/spikingjelly/datasets - Spike visualization
SNN_conversion_QCFS/spikingjelly/visualizing
Data and models
- SNN architectures
models - Spiking layers
models/layers.py - Dataset loading
data_loaders.py - Training utilities
utils.py
Other components
- Researcher
Connections 17
- Researcher → SNN training: configures training
- SNN training → Dataset loading: loads batches
- SNN training → SNN architectures: selects model
- SNN architectures → Spiking layers: uses layers
- SNN training → Training utilities: uses utilities
- Researcher → Train and evaluate: selects action
- Train and evaluate → Data preprocessing: prepares data
- Train and evaluate → ANN architectures: selects model
- Train and evaluate → ImageNet training: dispatches training
- Train and evaluate → ImageNet evaluation: dispatches evaluation
- Train and evaluate → Training and evaluation: calls workflows
- Train and evaluate → Activation conversion: replaces activations
- Train and evaluate → ANN-to-SNN conversion: converts model
- ANN-to-SNN conversion → Spiking neurons: uses neurons
- Spiking neurons → Neuron kernels: uses kernels
- Neuromorphic datasets → Spike encoding: provides event data
- Researcher → Spike visualization: inspects activity