Computational Photography, an AI-powered Slopendium

Book Notes Slides P-sets Syllabusdraft · v0.2.49.1

cover concept

cover concept · art via Reve and Seedream · loops animated with Kling · reload or click for another · animated

Computational Photography, an AI-powered Slopendium

by Frédo Durand, MIT

« Je n'ai fait celle-ci plus longue que parce que je n'ai pas eu le loisir de la faire plus courte. »

“If I had more time, I would have written you a shorter letter.”

Blaise Pascal

This is a very preliminary draft, and even the organization is still in flux. Feedback is welcome on content and organization, but fine-grained critique (wording, typography, line-level polish) is probably premature.

To leave feedback, select any text on a page and click the note / highlight button that pops up, or open the annotation panel with the tab at the top-right (‹). Notes are visible only inside our private review group.

Preliminary draft · v0.2.49

showopen inreading

2 FUNDAMENTALS OF IMAGING

3 VISUAL PERCEPTION AND COLOR

4 COLOR SPACES AND COLOR TECHNOLOGY

5.1The camera as an instrument Frédo approved
1. 5.1.1UI and display
2. 5.1.2Anatomy of a modern full-frame interchangeable-lens camera
3. 5.1.3Anatomy of cell-phone cameras
4. 5.1.4Camera versus phone
5. 5.1.5Types of cameras
6. 5.1.6Cameras beyond photography
7. 5.1.7A camera's other sensors
5.2Exposure and capture timing Frédo approved
5.3Viewpoint, lenses, and focus Frédo approved
1. 5.3.1Lenses and focal length
2. 5.3.2Aperture and the diaphragm
3. 5.3.3Focusing the lens
4. 5.3.4Choosing and checking autofocus
5. 5.3.5Image stabilization
6. 5.3.6Keeping the lens clean
7. 5.3.7Lens filters: polarizers, ND, and graduated ND
5.4Working with light Frédo approved
1. 5.4.1Natural illumination
2. 5.4.2Flash
3. 5.4.3Flash metering
4. 5.4.4Indirect illumination and bounce flash
5. 5.4.5Multiple-point lighting
5.5Developing and editing a photograph Frédo approved
1. 5.5.1The traditional darkroom
2. 5.5.2The digital darkroom
3. 5.5.3What a photographer actually does with those sliders
5.6Preparing photographs for viewing Frédo approved
5.7Photographic situations and tradeoffs Frédo approved
1. 5.7.1A moving person indoors
2. 5.7.2A backlit portrait
3. 5.7.3An interior with a bright window
4. 5.7.4A close-up object
5. 5.7.5A phone document capture
6. 5.7.6Adapting the cases: wildlife
7. 5.7.7Adapting the cases: night and sky
8. 5.7.8When the intended result is abstract
9. 5.7.9When the apparatus changes the problem
10. 5.7.10One frame, or a body of work
5.8Photographic representation and its limits Frédo approved
5.9Photography TLDR

6 LINEAR METHODS AND FOURIER ANALYSIS

7 PROBABILISTIC APPROACHES AND MACHINE LEARNING

8 BASIC IMAGE PROCESSING AND ISP

8.1Image representation Frédo approved
1. 8.1.1An image is an array
2. 8.1.2Float vs 8-bit vs more bits
3. 8.1.3Stochastic quantization and dithering
4. 8.1.4How the array sits in memory
5. 8.1.5Pixel coordinate conventions
6. 8.1.6Alpha and extra channels
7. 8.1.7Video and more dimensions
8. 8.1.8Pixel lookups and boundaries
9. 8.1.9Three kinds of operation
8.2What the numbers mean Frédo approved
1. 8.2.1What the numbers actually mean: encoding
2. 8.2.2What the numbers actually mean: color spaces
8.3Tone point operations Frédo approved
1. 8.3.1Three kinds of operation
2. 8.3.2The point-operation curve
3. 8.3.3Exposure: a multiply in linear light
4. 8.3.4Brightness vs. exposure
5. 8.3.5Contrast: steepen about a pivot
6. 8.3.6Black point, white point, and levels
7. 8.3.7The general tone curve
8. 8.3.8Lookup tables
9. 8.3.9Local point operations
8.4Beauty curves: the camera "look" Frédo approved
1. 8.4.1Why scene-linear looks dull
2. 8.4.2The shape of the curve
3. 8.4.3Highlights: roll off to white, not to a color
4. 8.4.4Saturation, and the rest of the "look"
5. 8.4.5Vignetting and grain, added on purpose
6. 8.4.6Our curve, in code
7. 8.4.7The discipline
8. 8.4.8Beauty curves in real ISPs
9. 8.4.9Perceptual basis for beauty curves
8.5Color point operations Frédo approved
1. 8.5.1Basic color enhancement: saturation and vibrance
2. 8.5.2The color edits, in code
3. 8.5.3Converting to black and white
4. 8.5.4Video color grading
8.6Color beautification and preferred colors Frédo approved
8.7Histograms Frédo approved
1. 8.7.1The histogram
2. 8.7.2The histogram depends on the encoding space
3. 8.7.3A histogram is a sampled estimate
4. 8.7.4Histogram equalization
5. 8.7.5Histogram matching
6. 8.7.6Constraining the slope: Ward's histogram-based tone mapping
8.8Sharpening Frédo approved
1. 8.8.1Linear sharpening
2. 8.8.2Why it's called "unsharp mask"
3. 8.8.3Sharpen in gamma, linear, or log?
4. 8.8.4Non-linear sharpening
8.9Dead pixels Frédo approved
1. 8.9.1Where they come from
2. 8.9.2Finding them
3. 8.9.3Fixing them
8.10Auto-exposure and auto white balance Frédo approved
1. 8.10.1Auto-exposure: metering
2. 8.10.2White balance and color constancy
3. 8.10.3White balance on film
4. 8.10.4Automatic white balance
8.11File formats and compression Frédo approved
1. 8.11.1The big picture: none, lossless, lossy
2. 8.11.2Data versus metadata: EXIF
3. 8.11.3PNG: the format we read from
4. 8.11.4JPEG: compression by perception
5. 8.11.5RAW files: before the cooking
6. 8.11.6HDR formats: more than 8 bits
7. 8.11.7Modern formats
8. 8.11.8Other formats: TIFF and GIF
8.12Recap ISP, non-destructive editing: Frédo approved
1. 8.12.1A basic ISP
2. 8.12.2Two stages this part did not teach
3. 8.12.3Real ISPs
4. 8.12.4Pipeline design and tuning
5. 8.12.5The ISP, evolving: traditional → learned → generative
6. 8.12.6Recap 2: non-destructive editing
8.13A raw developer in the browser

9.1Developing, testing and debugging Frédo approved
1. 9.1.1The workflow: build, run, look at the picture
2. 9.1.2Principles
3. 9.1.3Test on inputs you can verify by hand
4. 9.1.4Finding good real test inputs
5. 9.1.5Per-algorithm debug recipes
6. 9.1.6Crashes and bounds
7. 9.1.7Vibe coding: writing image code with an LLM
9.2The expert's eye as evaluation Frédo approved
1. 9.2.1Judge it yourself first
2. 9.2.2You are the expert, in a narrow sense
3. 9.2.3Training your eye
4. 9.2.4The input set decides what you can see
5. 9.2.5Beware the average, look at the worst case
6. 9.2.6Knowing when to stop, and when to ask
9.3Image metrics Frédo approved
1. 9.3.1Full-reference vs. no-reference
2. 9.3.2Why not just L2?
3. 9.3.3PSNR
4. 9.3.4SSIM
5. 9.3.5VDP and HDR-VDP
6. 9.3.6Learned metrics
7. 9.3.7The right metric depends on the task
9.4Towards meaningful user studies Frédo approved
1. 9.4.1How we got here
2. 9.4.2Should you run a study at all?
3. 9.4.3Three kinds of user research
4. 9.4.4Comparing outputs, and the checklist that makes it mean something
5. 9.4.5When the tool is the contribution
6. 9.4.6Asking before building
7. 9.4.7The rest of the toolbox
8. 9.4.8Reporting, and calling it what it is
9. 9.4.9Doing it ethically
10. 9.4.10Why these studies fail to replicate

10 EDGES MATTER

10.1Bilateral filtering Frédo approved
1. 10.1.1Bilateral filter
2. 10.1.2Robustness
3. 10.1.3Cross / joint bilateral
4. 10.1.4Bilateral grid
5. 10.1.5Bilateral upsampling and learning
6. 10.1.6Anisotropic diffusion and robust estimators
10.2Non-local means
10.3Edge-preserving optimization — colorization Frédo approved
10.4Local Laplacian filters Frédo approved
10.5Matting Laplacian Frédo approved
10.6Guided image filtering Frédo approved
1. 10.6.1The key idea: the output is a local line in the guide
2. 10.6.2Fitting the line, and the one knob
3. 10.6.3Why it is fast: O(N) regardless of window size
4. 10.6.4No gradient reversal
5. 10.6.5Reading the affinity back out
6. 10.6.6Where it is used
10.7Locally adaptive regression kernel (LARK) Frédo approved
1. 10.7.1Change the verb: from averaging to fitting
2. 10.7.2Steering the kernel to the local structure
3. 10.7.3The regression flavor of the affinity
4. 10.7.4LARK as a structure descriptor
10.8Poisson image editing Frédo approved
1. 10.8.1Edit gradients, not pixels
2. 10.8.2Seamless cloning: the headline application
3. 10.8.3Reconstruction = solving the Poisson equation
4. 10.8.4Poisson vs. pyramid blending
5. 10.8.5Advanced techniques
10.9Seam optimization Frédo approved
10.10Recap: which edge-aware technique when? Frédo approved
1. 10.10.1The three relationships to edges
2. 10.10.2Pros, cons, and cost: when to reach for each
3. 10.10.3A worked chooser

11 WARPING, MORPHING, AND LAGRANGIAN APPROACHES

12 MATCHING PIXELS AND HUMANS ACROSS SPACE AND TIME

13.1Noise models, transforms, and benchmarks Frédo approved
1. 13.1.1The noise that is actually there
2. 13.1.2Choosing the working domain
3. 13.1.3Variance-stabilizing transforms
4. 13.1.4A ladder of difficulty
5. 13.1.5Salt and pepper is a different problem
6. 13.1.6Benchmarks, and what a leaderboard number means
13.2Denoising by averaging Frédo approved
13.3Gaussian, median, and bilateral filtering Frédo approved
13.4Pyramid and wavelet denoising Frédo approved
1. 13.4.1The method: shrink the bands of a pyramid
2. 13.4.2Sparse in the right basis
3. 13.4.3Thresholding, hard and soft
4. 13.4.4Shrinkage as a MAP estimate
5. 13.4.5Ringing, and why the transform is usually overcomplete
6. 13.4.6Coring band by band, in practice
7. 13.4.7Coring and the Gaussian scale mixture
13.5Total variation and optimization-based denoising Frédo approved
1. 13.5.1Why the penalty's shape decides the answer
2. 13.5.2Isotropic, anisotropic, and the staircase
3. 13.5.3Solving it
4. 13.5.4Where it stands now
13.6Non-local means and BM3D Frédo approved
1. 13.6.1A photograph repeats itself
2. 13.6.2Non-local means: similarity as the weight
3. 13.6.3BM3D: group, transform, shrink, aggregate
4. 13.6.4The ceiling of hand-designed priors
13.7Learned denoisers Frédo approved
13.8Denoising without clean data Frédo approved
13.9Generative denoising with diffusion models Frédo approved
1. 13.9.1Why the posterior mean can blur detail
2. 13.9.2Sampling instead of averaging
3. 13.9.3The reverse process is a sequence of denoisings
4. 13.9.4What a hallucinated detail costs
13.10Video denoising Frédo approved
1. 13.10.1Temporal redundancy is the resource
2. 13.10.2Patches in space and time
3. 13.10.3Learned video denoisers
4. 13.10.4Temporal consistency, and the metric that does not measure it
13.11Denoising as a universal prior Frédo approved
13.12Measuring a denoiser Frédo approved
1. 13.12.1The reference metrics, and what each rewards
2. 13.12.2Comparing operating points
3. 13.12.3What to hold fixed
4. 13.12.4Ablations, cost, and what to report
5. 13.12.5The question a benchmark does not answer
13.13The limits of denoising
13.14Denoising TLDR

14 SUPER-RESOLUTION AND UPSCALING

14.1What can super-resolution actually do? Frédo approved
14.2Multi-frame super-resolution Frédo approved
1. 14.2.1Sub-pixel shifts are extra samples
2. 14.2.2The classical formulations
3. 14.2.3Registration is the hard part
4. 14.2.4What actually ships
5. 14.2.5The gain, and where it stops
14.3Learned burst methods Frédo approved
14.4Single-image super-resolution: classical methods Frédo approved
14.5Single-image super-resolution: deep learning Frédo approved
1. 14.5.1The architectural line
2. 14.5.2Where the upsampling goes
3. 14.5.3The ceiling, and what it was made of
4. 14.5.4What these networks do and do not learn
14.6The perception-distortion tradeoff and perceptual super-resolution Frédo approved
1. 14.6.1The tradeoff, stated
2. 14.6.2Buying perceptual quality: perceptual losses and adversaries
3. 14.6.3Measuring on both axes
4. 14.6.4The hallucination problem
14.7Generative super-resolution Frédo approved
1. 14.7.1Conditional generation
2. 14.7.2Sampling against averaging
3. 14.7.3Consistency, faithfulness, and what to report
4. 14.7.4The circle closes
14.8Reference-based and domain-specific super-resolution Frédo approved
1. 14.8.1Reference-based super-resolution
2. 14.8.2Face and text restoration
3. 14.8.3One model, any scale
4. 14.8.4Learning from the test image alone
5. 14.8.5The ordering, stated
14.9Blind and real-world super-resolution Frédo approved
14.10Video super-resolution and practical systems Frédo approved
1. 14.10.1Temporal consistency is the new constraint
2. 14.10.2Using the neighbors
3. 14.10.3Hybrid space-time super-resolution
4. 14.10.4What is deployed
5. 14.10.5What super-resolution actually delivers
14.11Super-resolution TLDR

15 DEMOSAICKING AND JOINT METHODS

15.1Basic traditional demosaicking Frédo approved
15.2Advanced traditional demosaicking Frédo approved
15.3Mosaics and aliasing Frédo approved
15.4Optimizing and searching classical demosaickers Frédo approved
1. 15.4.1Fitting a classical demosaicker by gradient descent
2. 15.4.2Searching for the program itself
15.5Alternative color filter arrays Frédo approved
1. 15.5.1Fuji's X-Trans
2. 15.5.2Quad-Bayer and its modes
3. 15.5.3RGBW and sparse color
4. 15.5.4Designing the array and the algorithm together
15.6Learned demosaicking Frédo approved
1. 15.6.1The data problem hiding inside a free training set
2. 15.6.2Architectures, briefly
3. 15.6.3Unrolling, and the bridge from the classical formulation
4. 15.6.4What the networks learned
5. 15.6.5Benchmarks and their age
15.7Neighboring problems in the raw stage Frédo approved
1. 15.7.1Defective pixels
2. 15.7.2Phase-detect autofocus pixels
3. 15.7.3Black level, clipping, and where pink highlights come from
4. 15.7.4Lens shading and crosstalk
5. 15.7.5The ordering claim, once more
15.8Joint demosaicking and denoising Frédo approved
1. 15.8.1The two orderings, made visible
2. 15.8.2The joint formulation
3. 15.8.3FlexISP: one energy, one prior
4. 15.8.4What it buys, and where it is deployed
5. 15.8.5Correcting chromatic aberration jointly
6. 15.8.6The argument generalizes
15.9Joint methods and the end-to-end pipeline Frédo approved
1. 15.9.1What end-to-end learning gains
2. 15.9.2What it costs
3. 15.9.3Keeping a useful intermediate representation
4. 15.9.4Where the field actually is
5. 15.9.5Where demosaicking sits in the pipeline
15.10Quad-Bayer and Nona-Bayer processing Frédo approved
1. 15.10.1Why the classical methods do not transfer
2. 15.10.2Remosaicking: making the sensor look like Bayer
3. 15.10.3Demosaicking the grouped layout directly
4. 15.10.4Remosaic or demosaic directly?
5. 15.10.5Split exposures within a block
15.11Synthetic data, unprocessing, and raw reconstruction Frédo approved
1. 15.11.1The inversion, step by step
2. 15.11.2Why an approximate inverse is good enough
3. 15.11.3Raw reconstruction as a goal of its own
4. 15.11.4The pipeline diagram, walked backwards
15.12Hardware alternatives to the mosaic Frédo approved
1. 15.12.1Pixel shift: spending time, precisely
2. 15.12.2Foveon and stacked photodiodes: spending silicon depth
3. 15.12.3Assorted pixels: spending spatial resolution on another axis
4. 15.12.4RGB-NIR: adding a band the eye does not have
5. 15.12.5Color wheels: spending time, sequentially, for many bands
6. 15.12.6Four ways to multiplex color
7. 15.12.7Routing color instead of filtering it
8. 15.12.8One table, one question
15.13Demosaicking TLDR

16.1What blur is, and what deblurring can undo Frédo approved
1. 16.1.1Why naive inversion fails
2. 16.1.2The fix, stated once
3. 16.1.3Which space to deblur in
16.2Non-blind deconvolution Frédo approved
16.3Priors for deconvolution Frédo approved
1. 16.3.1Sparse gradients: the prior that keeps edges sharp
2. 16.3.2Solving it, and the trick that made it fast
16.4Blind deblurring Frédo approved
1. 16.4.1Estimating the kernel and the image at once
2. 16.4.2Can you measure the kernel instead?
3. 16.4.3Why the obvious optimizer fails
4. 16.4.4Two ways to solve it
16.5Spatially varying blur Frédo approved
1. 16.5.1Rotational camera shake
2. 16.5.2A table of blurs, and which are convolutions
3. 16.5.3What ships in a phone
16.6Defocus, and undoing what the lens softened Frédo approved
16.7Learned deblurring Frédo approved
1. 16.7.1Networks that skip the kernel
2. 16.7.2The datasets are the argument
3. 16.7.3What learning did and did not change
16.8Video, bursts, and lucky imaging Frédo approved
1. 16.8.1Atmospheric seeing: why one long exposure fails
2. 16.8.2Shoot thousands, keep the sharpest
3. 16.8.3Align and stack the survivors
4. 16.8.4A poor man's adaptive optics
16.9Measuring a deblur Frédo approved
1. 16.9.1What went wrong, specifically
2. 16.9.2What to measure
3. 16.9.3On what data
4. 16.9.4What to report
16.10Deblurring TLDR

17 HIGH DYNAMIC RANGE CAPTURE

18 TONE MAPPING AND DISPLAY

18.1Tone mapping and perception Frédo approved
1. 18.1.1Why a faithful copy looks flat: four causes
2. 18.1.2The four do not multiply
3. 18.1.3What local tone mapping relies on
18.2Global tone mapping Frédo approved
1. 18.2.1Two families: global and local
2. 18.2.2Scale factors and logarithms: Ward and Drago
3. 18.2.3One curve for everything: Reinhard's photographic operator
4. 18.2.4Histogram adjustment: a curve shaped by the image
5. 18.2.5Why log space helps
6. 18.2.6From luminance to color
7. 18.2.7Why a single curve cannot win
8. 18.2.8Choosing and checking a tone map
18.3Two-scale and bilateral tone mapping Frédo approved
1. 18.3.1Starting with global tone mapping
2. 18.3.2Base and detail
3. 18.3.3LCIS: diffusion that stops at edges
4. 18.3.4Bilateral tone mapping
5. 18.3.5The decomposition zoo, and what it fixes
18.4Exposure fusion Frédo approved
1. 18.4.1Three qualities, one weight
2. 18.4.2Blending without seams
3. 18.4.3Why it survives
18.5Local Laplacian filtering and other multiscale methods Frédo approved
1. 18.5.1Multiscale models of adaptation
2. 18.5.2Subband companding: Li, Sharan and Adelson
3. 18.5.3Local Laplacian filters
4. 18.5.4Pyramids in exposure fusion and phones
18.6Gradient-domain tone mapping Frédo approved
18.7Tone mapping as perceptual optimization Frédo approved
1. 18.7.1Display-adaptive tone mapping
2. 18.7.2iCAM06: the color appearance route
3. 18.7.3Why they did not take over
18.8Display rendering transforms and color Frédo approved
1. 18.8.1From operators to view transforms
2. 18.8.2Per-channel curves and the hue problem
3. 18.8.3The responses: AgX, OpenDRT, ACES 2.0 and PBR Neutral
4. 18.8.4Looks and grading
5. 18.8.5Learned tone mapping, in brief
18.9High dynamic range in traditional photography Frédo approved
1. 18.9.1Three places to intervene
2. 18.9.2Combination printing: Le Gray and his successors
3. 18.9.3Reducing the range of the scene: lighting
4. 18.9.4Filters at capture: graduated neutral density and its relatives
5. 18.9.5Exposure and development: the negative's curve
6. 18.9.6The Zone System
7. 18.9.7Printing: paper grade, dodging and burning
8. 18.9.8Masking: the darkroom's base and detail
9. 18.9.9What carried over to computation
18.10HDR displays and display mapping Frédo approved
1. 18.10.1What a display can actually show
2. 18.10.2HDR displays and dual modulation
3. 18.10.3Display mapping: rendering for the display at hand
4. 18.10.4Gain maps on displays of different headroom
5. 18.10.5Inverse tone mapping
18.11Evaluating tone mapping Frédo approved
1. 18.11.1Why the comparisons disagree
2. 18.11.2What can be measured without a reference
3. 18.11.3What needs people
4. 18.11.4What to report
18.12HDR TLDR

19 LAYERS, TRANSPARENCY, COMPOSITING, AND MATTING

19.1Transparency and compositing Frédo approved
1. 19.1.1Compositing and alpha channels
2. 19.1.2Harmonization and where the blends live
3. 19.1.3Optical effects beyond alpha
19.2Segmentation Frédo approved
1. 19.2.1Segmentation: cutting the object out
19.3Blue and green screen matting, chroma key Frédo approved
1. 19.3.1Blue/green-screen matting and chroma keying
19.4Matting Frédo approved
1. 19.4.1The fundamental matting equation
2. 19.4.2Traditional matting approaches
3. 19.4.3Deep-learning matting
4. 19.4.4Generative matting

20 FIELD OF VIEW EXTENSION AND PANORAMA STITCHING

22 COMPOUND LENSES, AND ABERRATION CORRECTION

22.1Aberrations and optical challenges Frédo approved
1. 22.1.1Taxonomy of challenges
2. 22.1.2Spherical aberration
3. 22.1.3Coma
4. 22.1.4Astigmatism
5. 22.1.5Field curvature
6. 22.1.6Chromatic aberration
7. 22.1.7Radial distortion
8. 22.1.8Wave effects and diffraction
9. 22.1.9Vignetting
10. 22.1.10Flare and coating
22.2Aberrations correction Frédo approved
1. 22.2.1The two families of cure
2. 22.2.2Correction in glass
3. 22.2.3Computational correction
4. 22.2.4Radial distortion correction
5. 22.2.5Aberration data in the DNG, and the limits of per-copy correction
22.3Measuring lens quality Frédo approved
1. 22.3.1The modulation transfer function (MTF)
2. 22.3.2Measuring MTF in practice
3. 22.3.3Spot diagrams and the PSF
4. 22.3.4MTF through focus and depth of focus
5. 22.3.5Field-dependent MTF: sagittal vs tangential
6. 22.3.6Color, geometry, and illumination measurements
7. 22.3.7Scalar quality summaries
22.4Lens optimization Frédo approved
22.5A short bestiary of classic designs Frédo approved
1. 22.5.1The lens as a system: cardinal points, pupils, f-number, T-stop
22.6Scaling laws in optics Frédo approved
22.7Special optics Frédo approved
1. 22.7.1Tilt-shift and the Scheimpflug principle
2. 22.7.2Fisheye and non-rectilinear projection
3. 22.7.3Mirrors: catadioptric and reflecting systems
4. 22.7.4Periscope / folded-lens design (smartphone telephoto)
5. 22.7.5Anamorphic optics
6. 22.7.6Stereo (3D) lenses
7. 22.7.7Macro, microscope objectives, and telescopes
8. 22.7.8Teleconverters
9. 22.7.9From shaped glass to thin structures: Fresnel, diffractive, GRIN, metalenses
22.8Focus Frédo approved
1. 22.8.1Focus mechanics
2. 22.8.2Focusing a compound lens: beyond unit focusing
3. 22.8.3Focus actuators
4. 22.8.4Focus stacking, macro, and focusing rails
5. 22.8.5Optional background: viewfinder optics
6. 22.8.6Manual-focus aids: making the focal plane visible
7. 22.8.7Optional background: focus control in cinema lenses
22.9Autofocus Frédo approved
1. 22.9.1Contrast-detection AF
2. 22.9.2Phase-detection AF (the split-pupil / stereo trick)
3. 22.9.3On-sensor PDAF and dual-pixel AF
4. 22.9.4Depth from focus / defocus, and learned subject AF
5. 22.9.5Where to focus: saliency, faces, and eyes
6. 22.9.6Focusing in astrophotography
22.10Glare suppression Frédo approved
1. 22.10.1Where stray light comes from: flare, ghosting, veiling glare
2. 22.10.2Hardware suppression
3. 22.10.3Computational deflare and glare deconvolution
22.11Optical stabilization Frédo approved
22.12The eye as an optical instrument: vision and its correction Frédo approved
1. 22.12.1The eye as a camera
2. 22.12.2Refractive errors: the eye out of focus
3. 22.12.3Measuring the eye's optics
4. 22.12.4Correcting vision
5. 22.12.5Presbyopia and the bifocal problem: from bifocals to AF glasses

23 DEPTH OF FIELD

24 ILLUMINATION AND REFLECTION

25 NPR AND STYLIZATION

26 AI FOR PHOTOGRAPHY

26.1Editing as a programoutline
1. 26.1.1The edit as a recipe
2. 26.1.2Learning a sequence of operations, without language
3. 26.1.3Predicting a curve against predicting a plan
4. 26.1.4When pixels are the right output
26.2Language as an editing interfaceoutline
1. 26.2.1What people say, and what tools do
2. 26.2.2From a phrase to an operation and an amount
3. 26.2.3Regions: "the sky", "her face"
4. 26.2.4Asking back
26.3Language models that call vision toolsoutline
1. 26.3.1A language model as a dispatcher
2. 26.3.2Writing a program over image modules
3. 26.3.3What these systems cannot see
26.4Planning a retouchoutline
1. 26.4.1Teaching a model to see an adjustment
2. 26.4.2A plan with reasons
3. 26.4.3An agent over a real editor's tools
4. 26.4.4Editable output
26.5Verify and correctoutline
1. 26.5.1Decompose, act, check
2. 26.5.2Who does the checking
3. 26.5.3Stopping
26.6AI editing in productsoutline
1. 26.6.1Driving the tools already there
2. 26.6.2Describing the result instead
3. 26.6.3Assistants that call an editor
26.7Trust, authorship and evaluationoutline
1. 26.7.1Did it do what was asked?
2. 26.7.2Can the user see what it did?
3. 26.7.3Does the photograph still say what it recorded?

28 PHOTO COLLECTIONS

28.1Photo mosaics Frédo approved
1. 28.1.1The tiling-and-matching pipeline
2. 28.1.2Color correction and avoiding repeats
3. 28.1.3Multi-scale and irregular tilings
4. 28.1.4Why it resolves into the target at a distance
28.2Retrieval Frédo approved
28.3Auto curation Frédo approved
1. 28.3.1Technical quality: the easy rejects
2. 28.3.2Aesthetics: the hard, learned part
3. 28.3.3Grouping, summary, and diversity
4. 28.3.4Arranging the collection after culling
28.4Life logging cameras Frédo approved
1. 28.4.1The devices and how they fire
2. 28.4.2The memory-prosthesis reframing
3. 28.4.3The big-data problem passive capture creates
4. 28.4.4Privacy and ethics
28.5Photo tourism Frédo approved
1. 28.5.1Structure-from-motion on internet collections
2. 28.5.2From reconstruction to experience
3. 28.5.3Lineage
28.6Photobios Frédo approved
1. 28.6.1Align and order the collection
2. 28.6.2Let the data fill the gaps
3. 28.6.3Collection as experience
28.7Average Explorer Frédo approved
1. 28.7.1Galton's composite portraiture: what it was for
2. 28.7.2Alignment is everything
3. 28.7.3The average of a category
4. 28.7.4An artistic lineage
5. 28.7.5AverageExplorer: averaging made interactive
6. 28.7.6What an average is, and what it is good for
28.8Pix 2 GPS Frédo approved
1. 28.8.1Geolocation as retrieval over a geotagged corpus
2. 28.8.2The answer is a distribution
3. 28.8.3Mapping the collection itself
4. 28.8.4The learned successors
28.9Personalized priors Frédo approved
1. 28.9.1Personalized restoration
2. 28.9.2Personalizing generative models
3. 28.9.3The bargain and its ethics
28.10Artistic projects with photo collections Frédo approved
1. 28.10.1Statistical collage: Salavon
2. 28.10.2Anticliché camera
28.11Pareidolia
28.12Displaying images together Frédo approved
1. 28.12.1Selection
2. 28.12.2Layout
3. 28.12.3Color and coherence
4. 28.12.4Themes

29 MAKING COMPUTATIONAL PHOTOGRAPHY CODE FAST

31 LIGHT FIELDS AND PLENOPTIC CAMERAS

32 MULTI-APERTURE IMAGING

32.1Camera arrays: one rig, many instruments Frédo approved
1. 32.1.1One rig, four instruments
2. 32.1.2Commercial arrays: the Light L16
3. 32.1.3Where this sits
32.2Bullet time
32.3Multi-camera phones

33 COMPUTATIONAL OPTICS AND CODED IMAGING

33.1Wavefront coding Frédo approved
1. 33.1.1Why you cannot just deblur defocus
2. 33.1.2The fix: re-engineer the blur
3. 33.1.3The cubic phase plate
4. 33.1.4The depth-invariant PSF and a single deconvolution
5. 33.1.5What it costs
6. 33.1.6Where it sits
33.2Compressive sensing Frédo approved
33.3Coded aperture Frédo approved
33.4Phase-coded apertures Frédo approved
1. 33.4.1A reminder on wavefront coding
2. 33.4.2Focus sweep
3. 33.4.3The lattice-focal lens
4. 33.4.4Where it sits
33.5Code in time (phase, amplitude) Frédo approved
33.6Theoretical analysis of imaging systems in the 4D light field Fourier domain Frédo approved
33.7End-to-end optimization Frédo approved
1. 33.7.1The pipeline as one differentiable graph
2. 33.7.2Backpropagating into the glass
3. 33.7.3What gets designed: a height map, not a hyperparameter
4. 33.7.4A gallery of deep-optics results
5. 33.7.5What it costs, and where it can go wrong
6. 33.7.6Where it sits
33.8Fourier optics Frédo approved
33.9Exotic / advanced opticsoutline
1. 33.9.1Lensless imaging
2. 33.9.2GRIN
3. 33.9.3Metalenses and advanced crazy optics à la Barbastathis (coherent though)
4. 33.9.4Non-linear optics
33.10Optical modulators (spatial light modulators): DMD and LCD/LCoSoutline

34 COMPUTATIONAL SENSORS

35 COMPUTATIONAL ILLUMINATION

35.1Flash photography Frédo approved
1. 35.1.1Flash / no flash
2. 35.1.2Ramesh's multiflash
3. 35.1.3Removing flash artifacts
4. 35.1.4Dark flash and covert infrared photography
35.2High-speed and stroboscopic photography Frédo approved
35.3Illumination-based matting Frédo approved
35.4Separation of direct and global illumination Frédo approved
1. 35.4.1The frequency insight
2. 35.4.2Nayar's program: programmable, structured illumination
3. 35.4.3Toward coherent separation
35.5Light domes Frédo approved
1. 35.5.1The reflectance field and one-light-at-a-time capture
2. 35.5.2Relighting by linear combination
3. 35.5.3Scaling down: tabletop LED domes
4. 35.5.4Scaling out: the dome taken into the wild
5. 35.5.5The found dome: the eye
35.6Automatic aesthetic lighting Frédo approved
1. 35.6.1Computational bounce flash
2. 35.6.2Drone lighting: flying the light into place
35.7Dual photography Frédo approved
35.8Coherent imaging Frédo approved

36 3D AND DEPTH

37 INTEGRAL AND IMMERSIVE IMAGING

37.1Stereo glasses Frédo approved
1. 37.1.1Wheatstone's stereoscope (1838)
2. 37.1.2Routing a different image to each eye
3. 37.1.3Shooting and synthesizing a stereo pair
37.2VR goggles Frédo approved
37.33D displays with accommodation Frédo approved
1. 37.3.1Four ways to deliver a focus cue
2. 37.3.2The other gaps, and the ultimate display
37.4Lenticular displays Frédo approved
37.5Holography Frédo approved
1. 37.5.1Lippmann and Gabor: recording the wave
2. 37.5.2Off-axis holography and the space-bandwidth wall
3. 37.5.3Computational holography
37.6Retinal projection Frédo approved

38 REVEALING THE INVISIBLE

39 ADJACENT FIELDS AND APPLICATIONS

39.1Optical computingoutline
39.2Astrooutline
1. 39.2.1Extreme long exposure
2. 39.2.2Tracking, stacking, and selection (pointers)
39.3X-rayoutline
39.4Microscopyoutline
39.5Mm-waveoutline
39.6Music, soundoutline
39.7Fluorescenceoutline
39.8Opto-acousticoutline
39.9Ultrasoundoutline
39.10Aerial imagingoutline
39.11Computer visionoutline
39.12Robotics, drivingoutline

40 HUMAN FACTORS

40.1Human factors and the art of photography Frédo approved
1. 40.1.1Make better photos
2. 40.1.2Typical shooting scenarios
3. 40.1.3Macro photography
4. 40.1.4Special effect photography
5. 40.1.5Fun artsy stuff
6. 40.1.6Perception of art
40.2Ethics of computational photography
40.3Computational models of perceptionoutline
1. 40.3.1Spatial (and spatio-temporal) vision
40.4User studiesoutline
40.5Accessibility: photography by and for blind usersoutline
1. 40.5.1Blind camera — capture without a sighted operator
40.6The social and personal practice of photography Frédo approved

41 IMAGE FORENSICS AND AUTHENTICATION

42.1Programmable and modular cameras Frédo approved
1. 42.1.1What a closed camera hides
2. 42.1.2The Frankencamera: a camera as a pipeline of requests
3. 42.1.3From FCam to camera2
4. 42.1.4Opening cameras that were not meant to be opened
5. 42.1.5Modular cameras
42.2Image processing libraries Frédo approved
1. 42.2.1What a library provides
2. 42.2.2Eager whole-image libraries
3. 42.2.3Demand-driven, streaming libraries
4. 42.2.4Separating the algorithm from its schedule
5. 42.2.5GPU and tensor libraries
6. 42.2.6Choosing
42.3Lightroom-style raw developers Frédo approved
1. 42.3.1The recipe and the catalog
2. 42.3.2The pipeline and its order
3. 42.3.3Case study: CompPhotoRoom
4. 42.3.4What a raw developer knows about cameras
42.4Photoshop-style editors Frédo approved
1. 42.4.1The layer stack
2. 42.4.2Tiles
3. 42.4.3Case study: gamut
4. 42.4.4Scheduling: pull, cache, fuse
5. 42.4.5What it buys
6. 42.4.6Editors in the browser
42.5Networking and image transport Frédo approved
1. 42.5.1Delivering the right image
2. 42.5.2Tiles and pyramids over the network
3. 42.5.3Where to compute
4. 42.5.4Sending a recipe instead of the image
42.6Photography programming on phones Frédo approved
1. 42.6.1Android: camera2 and CameraX
2. 42.6.2iOS: AVFoundation, Core Image, and the photo library
3. 42.6.3The closed middle
4. 42.6.4Running the computation on the device

43 CONCLUSIONS, DISCUSSION

43.1Recap in contextoutline
1. 43.1.1Modern phones, multiple apertures, pano, HDR+
2. 43.1.2Recap: a modern mirrorless camera
3. 43.1.3Recap: a modern cell phone multi camera
4. 43.1.4Lightroom
5. 43.1.5Photoshop
43.2Why phones are so good (at photography) Frédo approved
1. 43.2.1Computation beats glass
2. 43.2.2The whole pipeline is co-designed
3. 43.2.3Machine learning and data at scale
4. 43.2.4Many small cameras for one big one
5. 43.2.5The human and system advantages
6. 43.2.6The hard caveats: physics still wins where it must
7. 43.2.7The throughline

45.1C++ for Python programmers Frédo approved
1. 45.1.1What changes, coming from Python
2. 45.1.2A first taste, side by side
3. 45.1.3A few more comparisons
4. 45.1.4Files and compilation
5. 45.1.5Value, reference, and `const`
6. 45.1.6Memory, briefly
7. 45.1.7Images in C++
8. 45.1.8Debugging, and the parts that hurt
9. 45.1.9Where to go next
45.2Refreshers Frédo approved
1. 45.2.1Linear algebra
2. 45.2.2Calculus: derivatives, gradients, integrals
3. 45.2.3Optimization and regression
4. 45.2.4Probability and information theory
5. 45.2.5Machine learning and deep learning
6. 45.2.6Programming: Python, C++, and PyTorch
45.3Problem Set 0 — Environment and C++ basics Frédo approved
1. 45.3.1Summary
2. 45.3.2Installation and Environment Setup
3. 45.3.3C++
4. 45.3.4Submission
45.4Problem Set 1 — Image class, point operations, and color Frédo approved
1. 45.4.1Summary
2. 45.4.2The Image Class
3. 45.4.3Brightness and Contrast
4. 45.4.4More Image Class Methods
5. 45.4.5Colorspaces
6. 45.4.6Spanish Castle Illusion
7. 45.4.7White Balance
8. 45.4.8Submission
45.5Problem Set 2 — Convolution and the bilateral filter Frédo approved
1. 45.5.1Summary
2. 45.5.2Smart Accessor
3. 45.5.3Blurring
4. 45.5.4Denoising using Bilateral Filtering
5. 45.5.5Extra credit
6. 45.5.6Submission
45.6Problem Set 3 — Denoising and demosaicking Frédo approved
1. 45.6.1Summary
2. 45.6.2Denoising from a sequence of images
3. 45.6.3Demosaicing
4. 45.6.4Edge-based green
5. 45.6.5Red and blue based on green
6. 45.6.66.8370 only (or 5% Extra Credit): Sergey Prokudin-Gorsky
7. 45.6.7Extra credit (maximum of 10%)
8. 45.6.8Submission
45.7Problem Set 4 — High dynamic range Frédo approved
1. 45.7.1Summary
2. 45.7.2HDR merging
3. 45.7.3Tone mapping
4. 45.7.4Extra credit (10% max)
5. 45.7.5Submission
45.8Problem Set 5 — Resampling, warping, and morphing Frédo approved
1. 45.8.1Summary
2. 45.8.2Resampling
3. 45.8.3Warping and morphing
4. 45.8.4Extra credit
5. 45.8.5Submission
45.9Problem Set 6 — Homographies and manual panoramas Frédo approved
1. 45.9.1Summary
2. 45.9.2Class morph
3. 45.9.3Homogeneous coordinates
4. 45.9.4Linear algebra
5. 45.9.5Warp and image with a homography
6. 45.9.6Compute homography from 4 pairs of points
7. 45.9.7Bounding boxes
8. 45.9.8Extra credit (up to 10% total)
45.10Problem Set 7 — Automatic panoramas Frédo approved
1. 45.10.1Summary
2. 45.10.2Previous problem set code
3. 45.10.3Class morph
4. 45.10.4Harris corner detection
5. 45.10.5Descriptor and correspondences
6. 45.10.6RANSAC
7. 45.10.7Automatic panorama stitching
8. 45.10.8Blending
9. 45.10.9Mini planet
10. 45.10.106.8370: Stitch N Images (6.8371: Extra Credit 5%)
11. 45.10.11Make your own panorama
12. 45.10.12Extra credits (10% max)
45.11Problem Set 8 — Non-photorealistic rendering Frédo approved
1. 45.11.1Summary
2. 45.11.2Paintbrush splatting
3. 45.11.3Painterly rendering
4. 45.11.4Oriented painterly rendering
5. 45.11.5Your image
6. 45.11.6Paper review (6.865 only)
7. 45.11.7Extra credits
45.12Problem Set 9 — Make-your-own, video, and ethics Frédo approved
1. 45.12.1Summary
2. 45.12.2Make your own assignment
3. 45.12.3Ethical issues in computational photography
4. 45.12.4Assignment lists
45.13EXIF and image metadata Frédo approved
1. 45.13.1What EXIF is
2. 45.13.2The fields, grouped by what they describe
3. 45.13.3How far to trust it
4. 45.13.4Privacy: the metadata that follows the picture
5. 45.13.5Reading and writing EXIF
45.14DNG: the Digital Negative Frédo approved
1. 45.14.1What DNG is, and the problem it solves
2. 45.14.2Inside the container
3. 45.14.3What the raw payload looks like: mosaic vs linear
4. 45.14.4The color recipe: matrices, profiles, and white balance
5. 45.14.5Opcodes: corrections the decoder must apply
6. 45.14.6Compression, and embedding the original
7. 45.14.7Where you meet DNG: adoption and relatives
8. 45.14.8Trade-offs, and DNG's relation to EXIF
45.15Rendering a raw DNG Frédo approved
1. 45.15.1The pipeline, step by step
2. 45.15.2Where Lightroom's "look" actually comes from
3. 45.15.3What we would ask an Adobe engineer to check
4. 45.15.4Two kinds of DNG, and the special cases
5. 45.15.5Easy mistakes (most of which we made)
45.16Datasets Frédo approved
1. 45.16.1Classification and features
2. 45.16.2Super-resolution
3. 45.16.3Deblurring and restoration
4. 45.16.4Denoising
5. 45.16.5HDR and tone mapping
6. 45.16.6Retouching and enhancement
7. 45.16.7Depth and motion
8. 45.16.8Light fields
9. 45.16.9Color and white balance
10. 45.16.10Faces
11. 45.16.11Inpainting, segmentation, and matting
12. 45.16.12Image quality
45.17A camera-feature wish list Frédo approved
1. 45.17.1Exposure, ISO, and dynamic range
2. 45.17.2Bracketing more than exposure
3. 45.17.3Focus and depth of field
4. 45.17.4Computational raw and the sensor
5. 45.17.5Motion data, metadata, and workflow
6. 45.17.6Panorama and multi-shot
7. 45.17.7The interface and the ecosystem
45.18How this book was created (the early process) Frédo approved
1. 45.18.1Two documents, not one
2. 45.18.2Compiling a section
3. 45.18.3Figures as code
4. 45.18.4Generative imagery: cover art and 3D
5. 45.18.5Verification and review
6. 45.18.6Keeping a long book coherent
7. 45.18.7The toolchain
8. 45.18.8What the machine did, and what it did not
9. 45.18.9Who wrote what: a per-part estimate
45.19How this book is made now Frédo approved
1. 45.19.1From compiling a book to editing one
2. 45.19.2How the author and an agent work together
3. 45.19.3Eleven agents and an ownership map
4. 45.19.4The files that hold it together
5. 45.19.5Skills: writing down what you repeat
6. 45.19.6The gates
7. 45.19.7What goes wrong now
8. 45.19.8What it cost
9. 45.19.9The bundle
45.20The course tutor: a local, book-grounded AI teaching assistant Frédo approved
1. 45.20.1What it is, and what it is for
2. 45.20.2Local-first
3. 45.20.3Grounded in the book: retrieval-augmented generation
4. 45.20.4How the retrieval works
5. 45.20.5Indexing other people's books
6. 45.20.6Measuring it, and keeping it from going stale
7. 45.20.7It links, it shows equations, it shows figures
8. 45.20.8Two front-ends, one core
9. 45.20.9What the instructor sees
10. 45.20.10Privacy and disclosure
45.21The semi-automatic grading system Frédo approved
1. 45.21.1The shape (to be confirmed)
2. 45.21.2Automatic versus human (to be confirmed)
3. 45.21.3To be filled in (from the instructor)
45.22Under the hood: prompts, patterns, and verifiers Frédo approved
1. 45.22.1Prompt patterns that made it work
2. 45.22.2The verifier suite
3. 45.22.3Why this is the interesting part
45.23Reading a Lytro light field Frédo approved
1. 45.23.1The container: the LFP/LFR format
2. 45.23.2The pipeline, step by step
3. 45.23.3From hexagonal lenslets to a uniform grid
4. 45.23.4Why a white image
5. 45.23.5Open questions and where we approximate
6. 45.23.6Easy mistakes (most of which we made)
45.24File conversion tools Frédo approved
1. 45.24.1JPG → PNG, without an alpha channel
2. 45.24.2DNG → a linear PNG
3. 45.24.3A Lytro capture → a Stanford-style light-field archive
4. 45.24.4How they run in the browser
5. 45.24.5Easy mistakes
45.25The interactive figures — a prompt-by-prompt making-of Frédo approved
1. 45.25.1Summary
2. 45.25.2Lorentz resonance: `fig-lorentz-resonance` (Figure 2.1.8)
3. 45.25.3Rainbow droplet: `fig-rainbow-droplet` (Figure 2.1.14)
4. 45.25.4Diffraction wave simulation: `fig-diffraction-wave-sim` (Figure 2.1.29)
5. 45.25.5Exposure-triangle simulator: `fig-exposure-triangle-sim` (Figure 2.11.5)
6. 45.25.6Repeated quantization: `fig-repeated-quantization` (Figure 3.1.4)
7. 45.25.7JPEG generation loss: `fig-jpeg-generation-loss` (Figure 3.17.9)
8. 45.25.8Exposure round-trip (JPEG): `fig-exposure-jpeg-roundtrip` (Figure 3.17.10)
9. 45.25.91-D sampling pipeline: `fig-sampling-1d-demo` (Figure 3.10.4)
10. 45.25.102-D sampling pipeline: `fig-sampling-2d-pipeline` (Figure 3.10.6)
11. 45.25.11Mitchell–Netravali bicubic (B,C): `fig-bicubic-bc` (Figure 3.11.13)
12. 45.25.12Rotation resample challenge: `fig-rotation-resample-interactive` (Figure 3.11.18)
13. 45.25.13Mini-Lightroom: `fig-mini-lightroom` (Figure 3.18.4)
14. 45.25.14Poisson blending: `fig-poisson-blend` (Figure 5.1.5)
15. 45.25.15Bilateral grid (3-D): `fig-bilateral-grid-3d` (Figure 5.2.15)
16. 45.25.16Perspective montage: `fig-perspective-montage` (Figure 6.6.13)
17. 45.25.17Beier–Neely morph: `fig-beier-neely-demo` (Figure 6.3.4)
18. 45.25.18Live face landmarks: `fig-face-landmarks-live` (Figure 7.10.2)
19. 45.25.19Lens optimizer: `fig-lens-optimizer-demo` (Figure 9.4.4)
20. 45.25.20Full auto-panorama: `fig-pano-stitch` (Figure 10.7.9)
21. 45.25.21CLIP-IQA curation: `fig-clip-iqa` (Figure 11.3.2)
22. 45.25.22Photobio time-lapse: `fig-photobio-demo` (Figure 11.7.2)
23. 45.25.23Refocus shift geometry: `fig-refocus-shift-geometry` (Figure 13.4.2)
24. 45.25.24Chroma key: `fig-chroma-key` (Figure 8.11.7)
25. 45.25.25Portrait-lighting simulator: `fig-portrait-lighting-sim` (Figure 2.11.26)
26. 45.25.26Aberration explorer: `fig-aberration-explorer` (Figure 9.1.6)
45.26Interactive demos index and search Frédo approved
1. 45.26.1Find a demo
2. 45.26.2Sharing, linking, and embedding a demo
3. 45.26.3Introduction
4. 45.26.4Fundamentals — light, optics, sensors
5. 45.26.5Visual perception and color
6. 45.26.6Color spaces and color technology
7. 45.26.7Photography
8. 45.26.8Linear methods and Fourier analysis
9. 45.26.9Probabilistic approaches and machine learning
10. 45.26.10Basic image processing and the ISP
11. 45.26.11Making computational-photography code fast
12. 45.26.12Edges matter
13. 45.26.13Warping and morphing
14. 45.26.14Matching pixels across space and time
15. 45.26.15Single-image computational photography
16. 45.26.16Compound lenses and aberration correction
17. 45.26.17Multiple-exposure imaging
18. 45.26.18Many images and photo collections
19. 45.26.19Video
20. 45.26.20Light fields and plenoptic cameras
21. 45.26.21Computational illumination
22. 45.26.223-D and depth
23. 45.26.23Revealing the invisible
24. 45.26.24Appendices and end matter
25. 45.26.25Not yet placed
45.27Course slides Frédo approved
1. 45.27.1The lectures
45.28Course slides, 2026 edition
45.29Course slides, earlier years Frédo approved
1. 45.29.12006, the first offering
2. 45.29.2The ENS / INRIA edition (around 2009)
3. 45.29.32011–2012, the Keynote editions
4. 45.29.4The MIT year-by-year editions (2007–2015)
45.30The Computational Photographer Frédo approved
1. 45.30.1Faculty jobs: why and how to get them
2. 45.30.2Can academia compete with the resources of industry?
3. 45.30.3Family in a box
4. 45.30.4Do good research, run a great lab
5. 45.30.5AI image synthesis is the biggest innovation in image making since photography
6. 45.30.6Lidar for autofocus
7. 45.30.7TedFest 2022
8. 45.30.8I want a light field camera
9. 45.30.9Lloyd Chambers's feature Christmas list
10. 45.30.10Ethics and computational photography
11. 45.30.11The DSLR will probably die. Are mirrorless the future of large standalone cameras?
12. 45.30.12Eurographics keynote slides
13. 45.30.13Halide talk
14. 45.30.14Do good research
15. 45.30.15Slides from my Siggraph talk about computational photography
16. 45.30.16No telephoto on cell phones
17. 45.30.17A perfect flat lens with metamaterials
18. 45.30.18Why video compression fails on confettis
19. 45.30.19This camera shows you a similar picture instead of the one you are taking
20. 45.30.20Project idea: automatic composite portrait
21. 45.30.21Lytro Cinema Camera: guestimates of characteristics
22. 45.30.22Auto white balance for flash and LEDs
23. 45.30.23Autofocus noise and microphone array
45.31How to do research Frédo approved
1. 45.31.1Frédo's own notes
2. 45.31.2A few more starting points
3. 45.31.3Further reading, talks, and advice
4. 45.31.4Odds and ends
45.32The Art and Science of Depiction Frédo approved
1. 45.32.1Introduction
2. 45.32.2There is no passive recording
3. 45.32.3What is a picture?
4. 45.32.4Introduction to human vision
5. 45.32.5Vision solves problems
6. 45.32.6Picture organization and Gestalt
7. 45.32.7Gaze movements and focal points
8. 45.32.8Color
9. 45.32.9Limitations and compensations I: the picture is flat
10. 45.32.10Limitations and compensations II: viewpoint and frame
11. 45.32.11The image is static; contrast and gamut are limited
12. 45.32.12Contrast
13. 45.32.13Representation systems
14. 45.32.14Perspective
15. 45.32.15Drawing and line drawing
16. 45.32.16Denotation systems
17. 45.32.17Tone system
18. 45.32.18Non-photorealistic rendering, and style
45.33Photoshop 1.0, running Frédo approved
1. 45.33.1What is on the menu bar
2. 45.33.2Running it in a browser
3. 45.33.3What to try
4. 45.33.4The source code is public, and that is not the same as buildable
5. 45.33.5How the program is put together
6. 45.33.6What the assembly relies on
45.34Photoshop 1.0 ported to JavaScript Frédo approved
1. 45.34.1Gaussian Blur, and five boxes
2. 45.34.2Unsharp Mask is thirty lines
3. 45.34.3Find Edges, without a square root
4. 45.34.4Add Noise, and the same theorem again
5. 45.34.5Median, and what happens at the border
6. 45.34.6Despeckle and Sharpen Edges are not kernels
7. 45.34.7Custom is the general case, and the built-ins are the special cases
8. 45.34.8On reading source code

46 MISSING STUFF, BUGS

46.1fig-learning-transition (hand-designed → learned → generative spectrum) — parked hereoutline
46.2Denoising by averaging is after basic denoisingoutline
46.3Photomosaics, my self organizing mapsoutline
46.4Rephotographyoutline
46.5Extreme low lightoutline
46.6Tilt shift,outline
46.7Misc.outline
46.8De-weathering (fog, rain)outline
46.9Near-infrared (NIR) photography *(relocated from REVEALING THE INVISIBLE, 2026-06-28)*outline