How bu-cisl/computational-miniature-mesoscope-cm2 works
Main System Components
1.1 Hardware Design (CAD & 3D Printing)
• CAD_models/CM2_V1 and CM2_V2 (LED housing, MLA housing, freeform illuminator, assembly files)
• 3D-printer (lab tabletop)
1.2 Optical Simulation Environment
• Zemax_models (ZOS, CFG files, spectra and coating profiles)
• Zemax ray-tracing engine
1.3 Data Acquisition & Storage
• Physical CM² device (sensor + LED array + MLA)
• Raw measurements (.tif, .mat)
• PSF libraries and sample data (downloaded or pre-rendered)
1.4 Reconstruction Pipelines
a) Model-Based 3D Deconvolution (MATLAB)
• Algorithm/model_based_3D_deconvolution_CM2_V1/cm2_related_code.m
• Utilities (ADMM_LSI_deconv_3D.m, bg_removal.m, etc.)
b) Deep-Learning CM²Net (Python/TensorFlow)
• Algorithm/cm2net/cm2_models.py
• Algorithm/cm2net/pred_cm2net.py
• Pretrained weights, sample measurement & ground truth (external drive)
Relationships & Data Flows
• Feed Zemax Models into two paths:
• Validate/design optics → Produce PSF/response data → Stored in PSF library
• Inform CAD tweaks (iterative design loop)
• Physical Device generates raw fluorescence measurements → Stored as .mat/.tif
• Reconstruction splits into two parallel pipelines:
• MATLAB Deconvolution: uses PSF library + raw data → 3D volume output
• CM²Net Inference: uses pretrained TensorFlow model + raw data → 3D volume output
• Both outputs feed a Visualization/Analysis block.
Architectural Patterns & Design Principles
• Modular, layered architecture: separate hardware, simulation, acquisition, and reconstruction.
• Two parallel algorithmic back-ends—model-based vs. data-driven—to allow comparison/benchmarking.
• Extensibility: new algorithms or hardware versions can plug into the same pipeline.
Call Out Key Technologies & Frameworks
• CAD: STL, OBJ, MTL (3D printing)
• Optical Simulation: Zemax (ZOS, CFG, spectra/coat profiles)
• Data Processing: MATLAB (R2020+, ADMM solver)
• Deep Learning: Python 3.7, TensorFlow 2.3
Detailed Layer Breakdown
• Layer 1: Hardware & CAD
• CAD_models (V1, V2) → 3D Printer → CM² Device
• Layer 2: Optical Simulation
• Zemax_models → PSF/response outputs → PSF Library
• Layer 3: Data Acquisition
• CM² Device captures fluorescence → Raw Data Store
• Layer 4: Reconstruction Engines (parallel)
• Model-Based: MATLAB scripts + PSF Library → Deconv Output
• Data-Driven: TensorFlow model + Raw Data → CM²Net Output
• Layer 5: Visualization & Downstream Analysis