Trial Period

Lab workshops for the trial period

A guide for tutors: the workshops that students who want to join HDSP complete during their trial period, with what each one covers and the evidence to request.

For tutors

What to request

Every workshop asks for the same kind of evidence. The specific figures and metrics are listed in each lab.

Code

A notebook (.ipynb) or scripts with executable code and intermediate figures.

Report

A short PDF with key figures, a small metrics table, and answers to the analysis questions.

Decisions and fixes

A short paragraph on what was changed during the lab and why.

Results

A results/ folder with the exported images.

Labs

Workshops

Download the full guide of each lab to share it with the student.

Optics

Image Formation and Spectral Acquisition

Guide dated June 11, 2026 · 21 pages

Students build a simple spectral imaging setup on the optical breadboard, acquire a 31-band reference cube with the monochromator and about 20 filter-wheel measurements, and reconstruct the spectrum.

Download PDF
Components Image formation Spectral GT (31 bands) Filters (N ≈ 20) Reconstruction Metrics

Learning goals

  • Validate focus on an optical rail using the thin lens model.
  • Assemble and align the test scene, lens, and monochrome camera.
  • Acquire the hyperspectral ground truth and the filter-based measurements.
  • Model acquisition as a linear system and reconstruct spectra (pseudoinverse, Tikhonov).

Deliverables

  • Report (2–4 pages) with 3–5 key figures, one metrics table, and analysis answers.
  • Capture log with exposure and gain per band.
  • results/ folder with exported figures and a README.txt.

Tools: Optical lab: monochrome camera, imaging lens, monochromator, filter wheel, broadband lamp

Optimization

Basic Optimization and Introduction to Inverse Problems

Guide dated July 22, 2026 · 13 pages

From calculus-based optimization in 1D and 2D to least squares and the pseudoinverse, then linear inverse problems y = Ax + e, deconvolution instability, and regularized reconstruction with ISTA.

Download PDF
Optimization 1D/2D Least squares Pseudoinverse Inverse problems Deconvolution ISTA

Learning goals

  • Read optimization notation and classify critical points with derivative and Hessian tests.
  • Derive the gradient of ‖Ax − y‖² and compare the pseudoinverse with lstsq.
  • Show why direct deconvolution is unstable under noise.
  • Implement ISTA with a sparse wavelet prior and study the effect of λ.

Deliverables

  • Written solution (PDF) with calculations, explanations, and answers.
  • Notebook or scripts with reproducible figures and PSNR/SSIM.
  • Short discussion on conditioning, priors, and reconstruction quality.

Tools: numpy, scipy, matplotlib, scikit-image, pywt

Image processing

Gamma Corrections, Geometric Transformations, and Representation in Another Basis

Guide dated July 29, 2026 · 7 pages

Students learn to apply each operation in the right domain: exposure in linear vs. gamma, affine warps with different interpolators, and DCT sparsification for compression. Plan 20–30 minutes per block.

Download PDF
Linear Exposure Gamma Warp DCT Metrics

Learning goals

  • Implement gamma correction and its inverse, and show its effect on histograms.
  • Apply affine transformations and compare nearest, bilinear, and bicubic interpolation.
  • Compress an image by keeping only the largest DCT coefficients.

Deliverables

  • Notebook (.ipynb) with conclusions per experiment and a decisions/fixes block.
  • Report (2–4 pages) with one figure per experiment and a PSNR/SSIM table.
  • Must use a personal image, not a standard course image.

Tools: Python 3.10+, numpy, matplotlib, scikit-image, scipy (opencv-python optional)

Deep learning

Deep Learning and Inverse Problems

Guide dated July 29, 2026 · 11 pages

From linear models to MLPs and CNNs in PyTorch, ending with a CNN autoencoder that restores blurred images. Every experiment runs on CPU; a GPU is optional.

Download PDF Starter notebook
Images Forward model Training Metrics

Learning goals

  • Describe inverse problems mathematically and recognize common cases.
  • Train an MLP for regression and classification.
  • Build and train a CNN autoencoder for image restoration.
  • Read loss curves, spot overfitting, and apply basic regularization.

Deliverables

  • Notebook (.ipynb) with conclusions per experiment and a decisions/fixes block.
  • Report (PDF) with one figure per experiment, metrics tables, and answers to all questions.
  • results/ folder inside a .zip file with exported images.

Tools: Python, PyTorch, the starter notebook (students work on their own copy)