MECH5720 Chap.9 Project Evidence and Exam Synthesis
Project Evidence and Exam Synthesis
The project and final examination both reward connected engineering arguments. A project claim should lead to a controlled comparison, calibrated measurement, reproducible inference and explicit validity range. An exam diagnosis should begin from the forward model, identify the broken assumption, predict an independent symptom and propose a discriminating test.
This chapter integrates optics, sampling, Fourier analysis, coded capture, lensless recovery and light fields while keeping observations separate from explanations. Project design starts with a statement that could be false. A claim about improved resolution should identify the frequency range, geometry and comparison rather than saying that an image looks better.
The baseline must challenge the claimed advantage: matched throughput for aperture comparison, fixed exposure and target for reconstruction comparison, or direct capture for relighting validation. Changing several factors at once prevents a causal conclusion. A run sheet records target distance, aperture, focus, exposure, gain, illumination, calibration and filenames.
Raw measurements should remain separate from corrected and reconstructed outputs. Analysis settings include normalisation, crop, boundary handling, parameter sweep and stopping rule. A versioned workflow that regenerates tables and figures reduces selective reporting and makes a repeat capture useful rather than confusing. Metrics need their own failure analysis.
Sharpness can increase with noise or ringing, a low residual can reflect overfitting, and a resolution target covers only known patterns. Combine measures that fail differently, such as measurement consistency, known spacing, repeatability and performance on a held-out target. Predefine the measurement region and aggregation rule before selecting the strongest result. Communication follows the evidence chain.
A figure caption should name condition, comparison and inference. A result paragraph separates observation from mechanism: first state what changed, then explain why the forward model predicts it, then give a control or uncertainty. Failures belong in the record when they establish where shift invariance, depth assumptions or sampling support stop holding. Exam reasoning uses the same discipline under time pressure.
Start with y = h*x+n, y = Mx+n or y = Ax+n as appropriate. Define coordinates and units, perform the calculation, interpret the trend and propose a discriminating test. A method name alone does not explain why an artifact occurred. Open-book preparation should be an index from symptom to model, assumption, equation and control rather than a long unstructured transcript.
The final examination is supervised, two hours and open book, carries 40 percent and has a written-exam hurdle. No reliable section pattern or question count is established, so revision should cover the connected concept sequence. Finish every answer at the strength of its evidence: distinguish what was measured, what was inferred, which alternative remains and the range over which the conclusion has actually been tested.
What this chapter covers
- 01
Falsifiable imaging claims
- 02
Controlled experiments
- 03
Calibration records
- 04
Metric failure modes
- 05
Bounded conclusions
- 06
Cross-topic exam diagnosis
Worked example · free
Audit a misleading sharpness claim
- +1Identify that the sharpness statistic can reward ringing and noise.
- +1Use the lower data residual as independent evidence for Method B.
- +1Use preservation of known target spacing as a second validation measure.
- +1Prefer Method B for this calibrated target while naming the tested conditions.
- +1Propose a repeated multi-frequency target test before extending the claim.
Key terms
- Control condition
- A comparison that isolates the effect of one experimental factor.
- Validation target
- A scene with known properties used to evaluate a method.
- Repeatability
- The stability of a result across repeated measurements under matched conditions.
- Bounded claim
- A conclusion limited to the range and conditions supported by evidence.
- Failure analysis
- A testable account of why a system departs from expected behaviour.
Project Evidence and Exam Synthesis FAQ
What belongs in a strong project comparison?
Use a falsifiable question, matched geometry and exposure, an appropriate baseline, recorded calibration and parameter settings, repeated measurements, a metric with stated units and an honest account of failures and validity limits.
How should an exam diagnosis be structured?
State the forward model, identify the assumption challenged by the observation, predict another symptom and propose a control that separates the diagnosis from the nearest alternative. Calculations should support that causal chain.
Why are attractive images insufficient evidence?
Visual quality can hide noise fitting, ringing, selective examples or incorrect geometry. Measurement residuals, known targets, repeat captures and parameter sensitivity provide checks with different failure modes and make the conclusion auditable.
Exam move
Turn each broad topic into a five-part card: model, assumption, calculation, failure symptom and control. Rehearse cross-topic questions that connect acquisition to reconstruction, such as how a PSF notch constrains deconvolution or how sampling changes a depth estimate. For project revision, practise rewriting broad claims with geometry, comparison, metric, uncertainty and validated range.
Audit one figure at a time: its caption should identify condition, comparison and inference without relying on surrounding prose. Pair each preferred metric with a way it can fail and a second measurement that responds differently. Under timed conditions, answer in four moves: write the forward model, identify the challenged assumption, compute or predict the consequence, and propose a discriminating test.
Close by bounding the conclusion rather than repeating the method name. Rehearse one two-minute explanation that connects a measured artifact to a controlled follow-up experiment.