Chapter 23 — Exercises

Computational photography is learned by running the machinery and judging the result — so most of these exercises send you out to shoot in difficult conditions and then ask you to look hard at what your phone did with the frames. A phone is the ideal (in fact the required) camera here. Difficulty is marked ⭐ (warm-up), ⭐⭐ (core), ⭐⭐⭐ (stretch). Do the starred and odd-numbered ones at minimum; model answers and critiques for those are in the book's Answers to Selected Exercises. Keep everything you shoot — your computational experiment feeds the Portfolio, and several of these "fail studies" are worth keeping as a record of what computation gets wrong.

A note before you start: whenever an exercise says "shoot it honestly first," it means take one ordinary single exposure with the computational mode off, as a baseline. The whole craft of this chapter is comparison — what did the computation add, and what did it fake?

Part A — Reading the machinery (Read the Frame)

Exercise 23.1 ⭐ — Spot the stack. Over one day, watch your own phone work. Find five situations where you can feel the multi-frame pause after pressing the shutter (dim room, backlit window, twilight) and five where the capture is instantaneous (bright daylight). Write one sentence per situation naming why the phone did or didn't stack. You are learning to sense, without a menu, when computation is engaging.

Exercise 23.2 ⭐⭐ — Diagnose the phone look. Find five of your own existing phone photos and, for each, name at least one computational fingerprint: over-blue sky, waxy "watercolor" skin or foliage, crunchy over-sharpening, a portrait-mode hair halo, or a flattened-grey HDR look. For each, name the pipeline stage (from Figure 23.6) responsible. The goal is to stop seeing "a photo" and start seeing "a series of processing decisions."

Exercise 23.3 ⭐⭐ — Read a night frame. Find a night photo you (or someone) shot in night mode. Write a full Described Photograph of it using the six fields (THE FRAME / THE LIGHT / THE MOMENT / THE CHOICES / THE EFFECT / THE LESSON). In THE CHOICES, specifically address: was anything moving, and did the stack handle it? Where are the highlights, and did the streetlights blow out?

Exercise 23.4 ⭐⭐⭐ — Find the depth-map seam. Collect three portrait-mode images (yours or freely-available samples). Zoom into the edge between subject and blurred background on each and write, precisely, where the depth map succeeded and where it failed: trace the hair, the shoulders, any hand or held object, any glasses. Rank the three by how convincing the fake is, and explain the ranking by what made each scene easy or hard to segment.

Part B — Multi-frame and HDR (Shoot This / Settings Drill)

Exercise 23.5 ⭐ — Hold still and let it finish. Shoot the same dim indoor scene twice: once jerking the phone away the instant you hear the shutter, once holding rock-steady until the thumbnail settles. Compare them at 100%. The difference is the whole "let the stack finish" lesson, in your own hands.

Exercise 23.6 ⭐⭐ — The three-way HDR test. Find a high-contrast scene from a fixed vantage (a window-lit room with the bright outdoors visible, or a building's dark side against a bright sky). Shoot it three ways: HDR off exposed for the highlights; HDR off exposed for the shadows; HDR on. Lay all three side by side. Mark where each single exposure clipped, and judge whether the HDR tone mapping looks natural or cooked. (This is the chapter's 📸 In the Field — do it formally and keep the most honest frame.)

Exercise 23.7 ⭐⭐ — Expose for the highlights, rescue the shadows. In a backlit scene (subject in front of a bright window), shoot HDR two ways: once tapping to expose for the dark subject, once tapping to expose for the bright window. Compare how natural the merge looks. Write why exposing for the highlights and letting HDR lift the shadows usually wins.

Exercise 23.8 ⭐⭐⭐ — Settings Drill: protect the mood. You're shooting a deliberately moody scene: a single lamp in a dark room, the rest falling to near-black, the kind of low-key image from Chapter 8. Auto-HDR wants to open every shadow. Write out exactly what you'd do — mode, HDR setting, where you'd tap to expose, and whether you'd shoot RAW — to keep the darkness. Then shoot it and confirm your plan worked.

Part C — Night mode and motion (Shoot This / Fix the Photo)

Exercise 23.9 ⭐ — Static night mode. Go to your city block at night location (or any dark, empty scene: a quiet street, a lit building, a still landscape). Brace the phone on a ledge and shoot one night-mode frame, exposing down a touch so the brightest light keeps its detail. This is night mode in its home territory — note how much more the photo shows than your eye could.

Exercise 23.10 ⭐⭐ — Break night mode on purpose. Shoot the same dark scene with one thing moving through it: a person walking across, a passing car, your own hand waved slowly. Examine how the stack rendered the motion — ghost, smear, partial transparency, disappearance. Write what the rejection step (Figure 23.3, step 3) appears to have done. Keep this as a "failure study"; it's genuinely instructive.

Exercise 23.11 ⭐⭐ — The freeze test. Photograph a willing person at night in night mode three ways: (1) telling them nothing; (2) telling them "hold completely still" for the whole capture; (3) having them step into a pool of light (under a streetlamp, near a lit window) so the phone uses a shorter capture. Compare the sharpness of the face across all three. Write which worked and why, in terms of the multi-second capture.

Exercise 23.12 ⭐⭐⭐ — Fix the Photo: the ghosted night shot. A described failed shot: "I shot my friend in front of a beautiful lit fountain at night in night mode. The fountain and building came out gorgeous and clean, but my friend is a blurry, half-transparent ghost." Diagnose precisely why (what stacked well, what didn't), then prescribe three different re-shoots that would fix the friend without losing the background. Then stage and shoot your own version to prove the fix.

Part D — Portrait mode and depth (Shoot This / Recreate It)

Exercise 23.13 ⭐⭐ — Distance is the fix. In your kitchen window location, shoot a subject (person, toy, plant) in portrait mode twice: once with the background right behind them, once with them stepped three metres forward. Zoom into the edges of both. Write how much real distance improved the synthetic-bokeh edge, and connect it to why a real lens would want the same separation.

Exercise 23.14 ⭐⭐ — Moderate beats maximum. Shoot one portrait-mode subject at your phone's blurriest simulated-aperture setting and again at a moderate one. Compare the believability of the background blur and the severity of any edge artifacts. Write why a moderate blur often looks more real than a maximal one. (This is the chapter's 📸 In the Field — extend it across all four artifact triggers if you can.)

Exercise 23.15 ⭐⭐⭐ — The hard-portrait gauntlet. Deliberately feed portrait mode its four hardest cases: flyaway/curly hair; eyeglasses; a hand or mug reaching toward the camera; a subject against a busy close background (a bookshelf, a fence). Shoot all four. Find and screenshot every artifact. Write one sentence per case on why the depth map struggled, and what (if anything) you could do to help it.

Exercise 23.16 ⭐⭐ — Recreate the optical look, two ways. If you have access to any dedicated camera with a lens that opens to f/2.8 or wider (even an old kit zoom at its widest), shoot a portrait with real optical blur. Then recreate the same composition in phone portrait mode. Compare the two background blurs honestly: where is the synthetic one convincing, and where does the real optical blur reveal it as a fake (the falloff with distance, the edges, the highlights)? If you have no such camera, recreate from a Described Photograph instead and reason it through in writing.

Part E — Color, sharpening, and RAW (Shoot This / Five Conditions)

Exercise 23.17 ⭐⭐ — JPEG vs. RAW, same frame. If your phone can shoot RAW (or RAW+JPEG), capture one scene as both. Open them side by side. Catalog every difference the computational pipeline introduced into the JPEG: color saturation, sky/skin treatment, contrast, sharpening, smoothing of fine texture. The RAW is your "before"; the JPEG is the phone's opinion. If your phone can't shoot RAW, document the JPEG's choices against what your eye saw at the scene.

Exercise 23.18 ⭐⭐ — Hunt the watercolor smear. Find or shoot a low-light phone photo containing fine texture (foliage, hair, distant brickwork, fabric weave). Zoom to 100% and find where noise reduction has smeared texture into a painterly mush. Then shoot the same texture in better light and confirm the smear lessens. Write the relationship between light, noise, noise-reduction, and the loss of fine detail.

Exercise 23.19 ⭐⭐⭐ — Five conditions, one face (or object), computation watched. Photograph one subject across five conditions — bright daylight, open shade, dim indoor, backlit window, and after dark — letting the phone do whatever it does automatically each time. For each, name which computational modes engaged and what they did to the subject (especially any skin smoothing, color push, HDR, or night-mode stacking). Self-select the two frames where computation served the subject and the one where it lied. Defend your picks.

Exercise 23.20 ⭐⭐ — Accurate color challenge. Find a scene where color accuracy matters: a brightly-colored product against a neutral wall, a paint chip, a piece of art, a food plate you want true-to-life. Shoot it in normal auto mode, then (if possible) in RAW with white balance you set yourself. Compare against the real object in your hand. Write how far the memory-color boost drifted from reality, and when that drift would be a problem.

Part F — Interleaved (mixing prior chapters)

Exercise 23.21 ⭐⭐ — Computation meets exposure (Ch.3). For one HDR scene, predict in advance — using the exposure triangle and the histogram idea from Chapter 3 — which parts of the scene a single exposure would clip (consult the highlights and shadows). Then shoot a single frame and check your prediction against the actual clipping. Then shoot HDR and see what it recovered. You're connecting "dynamic range" the computational concept to "exposure" the manual one.

Exercise 23.22 ⭐⭐ — Computation meets light direction (Ch.5). Shoot a portrait-mode subject lit two ways: flat front light, and soft side light from a window (Chapter 5's direction lesson). Notice whether the light direction affects how well portrait mode segments the edge — does the shadow side of the face confuse the depth map? Write what you find.

Exercise 23.23 ⭐⭐⭐ — Computation meets the moment (Ch.10 / Ch.21). Night mode and the decisive moment (Chapter 10) are in tension — the multi-second capture can't freeze a peak instant. Go to a night scene with occasional motion (a quiet street with the rare passer-by). Plan how to get a clean night-mode background and a frozen subject moment: would you composite in your head across two captures, time the capture for a still beat, or add light for a short capture? Shoot your best attempt and write what you'd refine. Connect it to the long-exposure motion choices from Chapter 21.

Exercise 23.24 ⭐⭐ — The override audit (Ch.2, Ch.22). Take five computational keepers from earlier in your portfolio or camera roll and, for each, decide in hindsight: should this have been shot RAW (Chapter 2) or with manual exposure (Chapter 22) instead of trusting the auto pipeline? Defend each call using the override table (Figure 23.7). This trains the single most valuable habit of the chapter — deciding when to take the controls.

Part G — Portfolio and reflection

Exercise 23.25 ⭐⭐ — The Portfolio computational experiment. Complete the Portfolio Checkpoint: reshoot a genuinely tricky scene (backlit, dark, or needing subject-background separation) with a computational mode, having first shot the honest single-exposure baseline. Keep the best computational frame and write the two required curation lines — what it added and what it faked.

Exercise 23.26 ⭐ — Light Log: notice computation in the wild. Add today's Light Log entry about a computed image you encountered — your own or in the world (an ad, a real-estate listing's too-perfect interior, a friend's over-smoothed portrait). What did the computation add, and what did it fake? You're extending the Light Log from "seeing light" to "seeing how light was rendered."

Exercise 23.27 ⭐⭐⭐ — Teach the override. Explain to someone — a friend, a journal, a camera pointed at yourself — when your phone's computational modes help and when they lie, using two of your own frames as evidence (one where computation rescued a shot, one where it faked or wrecked something). Teaching forces clarity; if you can teach the override decision, you own it.

Stretch challenge

Exercise 23.28 ⭐⭐⭐ — One scene, the full computational map. Choose one ambitious scene that exercises every mode — say, a dim interior at dusk with a person, a bright window behind, and texture worth keeping. Shoot it: (1) honest single exposure; (2) HDR; (3) night mode; (4) portrait mode on the person; (5) RAW. Then write a one-page report mapping each capture to what the computation did — what each added, what each faked, and which capture you'd actually keep and why. This single exercise integrates the whole chapter and is the best possible preparation for the digital darkroom in Part VI, where you'll develop that RAW yourself.

Model answers and critiques for the starred and odd-numbered exercises appear in the book's "Answers to Selected Exercises." Run each with your phone before reading them — in computational photography, watching the machinery work (and fail) on a real scene teaches more than any explanation.