Performance Metrics for Neuromorphic Imaging

Simulation and analysis code for: Performance metrics for neuromorphic imaging paper
N. Kruger, S. Arja, E. Andrew, T. Monk & A. van Schaik (SPIE 13376, 2025)
▶ Read the full paper on SPIE Digital Library

Summary

This is a framework to simulate basic object motion scenes, as viewed by an event camera, and calculate the expected performance of the sensor for various imaging system parameters.

This repository provides:

1. Config-driven scene setup via simple .ini files (sensor, optics, scene).
2. Synthetic frame generation based on those parameters.
3. Event‐stream simulation with classification of target vs. background events.
4. Performance analysis scripts to extract metrics from the event data.

---

Setup

Requirements

• python: 3.9.x, 3.10.x

Tested environments

• Ubuntu 22.04 and Windows 11
• Conda 23.1.0
• Python 3.9.18

Installation

git clone https://github.com/neuromorphicsystems/dvs_performance_metrics.git
cd dvs_performance_metrics
conda env create -f environment.yml
source ~/.bashrc && conda activate dvs_performance_metric
python3 -m pip install -e .

Run Simulation

Follow these steps to configure and launch your simulations:

1. Prepare your configuration files

• Create one or more <name>.ini files in the config/ folder.
• Each .ini contains your imaging parameters (sensor, optics, scene, timing).
• See examples in config/

2. Launch the simulator

Single‐config-run

python3 run_simulation.py -filename "<config_name>" #no need to add the file format ".ini" for config filename

# Example command
python3 run_simulation.py -filename "frequency_size_heatmap_size50"

Optional flags

Edit the top of run_simulation.py

DO_PLOTS    = 0    # 1 = enable real-time matplotlib display
SAVE_FRAMES = 0    # 1 = save per-frame PNGs into OUTPUT/<config>/
epoch       = 5    # number of repeats per config
bgnp, bgnn  = 0.2, 0.2  # background event rates
blankEvRun  = 0.5  # seconds of noise “warm-up”

3. Data Analyis and Post-processing

When all data files are created, use analysis scripts to examine the data and calculate change in various metrics. Examples include MATLAB scripts such as FullTestAnalysis.m, and these need to be adapted to the parameter of interest of each simulation run.

Content of each folder

• OPTICAL_SIMULATOR/
  Core routines for generating synthetic image frames.

  • basic_tar_bg_simulation.py:
    • Parses your .ini config.
    • Creates target and background images based on scene, optics, and sensor parameters.

• EVENT_SIMULATOR/
  Event‐stream generation engine (forked and extended from Joubert et al. 2021):
  https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2021.702765/full

  • Adds physical-to-voltage conversion (illumination → pixel voltage).
  • Labels each event as signal, background, or noise.

• PERFORMANCE_METRICS/
  Utility functions that support the MATLAB analysis scripts (e.g. motion alignment, metric calculators).

• dvs_warping_package/ & dvs_warping_package_extension/
  Python and C++ libraries for:

  • Event denoising
  • Motion compensation (event “warping”)
  • Only used when needed

• config/
  Your simulation .ini files (scene, optics, sensor, noise, scan parameters).

OUTPUT/: Simulation data output files

OUTPUT/
└─ <config_name>/
   ├─ events_and_labels/
   │  ├─ simdata_<sim_name>_<ep>.mat    # MATLAB struct “simulation_data”
   │  └─ ev_<sim_name>_<ep>.txt         # [x, y, p, ts, l] per event
   ├─ raw_event_image/        (if SAVE_FRAMES=1)
   ├─ only_signal_image/      (if SAVE_FRAMES=1)
   ├─ only_background_image/  (if SAVE_FRAMES=1)
   ├─ only_noise_image/       (if SAVE_FRAMES=1)
   ├─ mask_overlay/           (if SAVE_FRAMES=1)
   └─ labeled_image/          (if SAVE_FRAMES=1)

Config file creation and structure

A few notes:

1. There are two type of config files - one to be called by for simulations, and the second for generic sensor configurations. We will only detail the simulation parameters config file here, as the sensor config file is only a sub-set of the former.

2. Each configuration file can include ONE parameter to be "scanned" by the simulator, meaning it can be included as a vector, and not a scalar value. The difference being that the values for the chosen scanned parameter are a row of values, seperated be a comma "," only. This can only be done on fields that are a scalar or integer inputs (not for text input).

3. Responsibility on ensuring legal values is on the operator of the simulator. Ensure you choose values that fit real-world scenarios and hardware parameters.

4. Brightness levels are those expected on the sensor plane. As this simulation doesn't include translating target illumination levels to these expected on the sensor plane, the analytical process of estimated precieved brightness by the sensor is left for the user to derive.

🚀 InitParams

General settings for the simulation.

Parameter	Type	Unit	Description
sim_name	string	—	Unique name for this run (also used to name output files).
t_end	float	seconds	End time of the simulation.
dt	float	seconds	Time increment for each simulation step.
obstruct	bool	—	True: target obscures the background; False: target and background additively combine. (Default: True.)
multiplier	int	—	Oversampling factor for pixel grid. (Only 1 has been fully tested.)
lux_flag	bool	—	True: use lumen units; False: use W/m² at a given wavelength.
wavelength	float	μm	Mean illumination wavelength (used only when lux_flag = False).
sensor_model	string	—	Which sensor to load: Gen4, Gen3, Davis346 or Manual. Non-Manual options pull from built-in config files. Manual uses the ManualSensorParams below.

---

🎬 SceneParams

Controls the geometry and motion of target & background.

Parameter	Type	Unit	Description
BG_const	float	lumen / W/m²	Constant background illumination level.
t_distance	float	meters	Nominal distance of the (flat) target from the imager.
bg_distance	float	meters	Nominal distance of the (flat) background from the imager.
t_velocity	float	m/s	Target speed (positive = rightward in LOS).
t_elevation	float	radians	Target elevation above the horizon.
t_azimuth	float	radians	Initial target azimuth (0 = North).
i_elevation	float	radians	Imager’s elevation angle—set equal to t_elevation to center the target in FOV.
i_azimuth	float	radians	Imager’s azimuth angle—set equal to t_azimuth to center the target in FOV at t = 0.
Jitter_amp	float	pixels	Amplitude of 1/f LOS jitter.
Jitter_speed	float	Hz	Base frequency multiplier for LOS vibration.
imaging_los_speed	float	rad/s	Initial angular velocity of LOS (overwritten in “perfect” tracking).
imaging_los_acc	float	rad/s²	LOS angular acceleration (used by “chase” & “leaps” heuristics).
tracking_mode	string	—	Heuristic: chase, leaps or perfect.
leapTime	float	seconds	Cycle time between “leaps” toward the target.
leapDuty	float	—	Duty cycle (fraction of leapTime) spent moving vs. stopping in “leaps” mode.

---

🔭 OpticParams

Defines the imaging optics.

Parameter	Type	Unit	Description
focal_length	float	meters	Lens focal length.
Fnum	float	—	f-number (focal_length ÷ aperture dia).
PSF_size	float	pixels	Gaussian PSF full-width at half max.

---

🎯 TargetParams

Shape, timing & brightness of the target.

Parameter	Type	Unit	Description
target_type	string	—	spot (constant) or g_flash (single-pulse).
target_radius	float	meters	Radius of the target disk.
target_brightness	float	lumen / W/m²	Peak brightness on the sensor.
t_init	float	seconds	Flash onset time (for g_flash).
t_constant	float	seconds	Flash duration (approx. FWHM).
target_brightness_min	float	lumen / W/m²	Min. brightness (blinking_spot & modulated_spot, TBD).
mod_freq	float	Hz	Modulation frequency (blinking_spot & modulated_spot, TBD).
mod_duty_cycle	float	—	Duty cycle of “on” time (blinking_spot, TBD).

---

🌄 BgParams

Background pattern settings.

Parameter	Type	Unit	Description
BG_type	string	—	const, lines or natural.
S_freq	float	lines/m	Spatial frequency (or characteristic scale for natural).
S_dir	float	degrees	Orientation of lines (0 = left–right, 90 = up–down).
BG_brightness	float	lumen / W/m²	Maximum brightness for lines or natural. (Set to 0 for constant BG.)

---

⚙️ SensorBiases

User-tunable thresholds & refractory time.

Parameter	Type	Unit	Description
diff_on	float	—	ON-event threshold.
diff_off	float	—	OFF-event threshold.
refr	float	μs	Refractory period.

---

🛠️ ManualSensorParams

Override built-in sensor specs when sensor_model = Manual.

Parameter	Type	Unit	Description
width	int	pixels	Horizontal resolution.
height	int	pixels	Vertical resolution.
pixel_pitch	float	meters	Pixel pitch.
fill_factor	float	—	Active-area fraction.
tau_sf	float	seconds	Source-follower time constant[1].
tau_dark	float	seconds	Dark-current time constant.
QE	float	—	Quantum efficiency (0–1).
threshold_noise	float	—	Reset-noise standard deviation.
latency_jitter	float	seconds	Jitter of event latency.
latency	float	seconds	Fixed additional latency (e.g., arbiter delay).
I_dark [1]	float	amps	Mean dark current (must be >0).

^[1] See “Re-interpreting the step-response probability curve to extract fundamental physical parameters of event-based vision sensors” (SPIE Proc. 10.1117/12.3022308)

Citations

If you use this work in an academic context, please cite the following:

@inproceedings{kruger2025performance,
  title        = {Performance metrics for neuromorphic imaging},
  author       = {Kruger, Nimrod and Arja, Sami and Andrew, Evie and Monk, Travis and van Schaik, André},
  booktitle    = {Quantum Sensing and Nano Electronics and Photonics XXI},
  volume       = {13376},
  pages        = {74--82},
  year         = {2025},
  organization = {SPIE}
}

@article{joubert2021event,
  title     = {Event camera simulator improvements via characterized parameters},
  author    = {Joubert, Damien and Marcireau, Alexandre and Ralph, Nic and Jolley, Andrew and Van Schaik, André and Cohen, Gregory},
  journal   = {Frontiers in Neuroscience},
  volume    = {15},
  pages     = {702765},
  year      = {2021},
  publisher = {Frontiers Media SA}
}