The Thread: notes on shopping an outfit from a photo

How it works

Why does a search return fewer items than you see?

You count nine garments in the photo and the results list four. The missing five were not skipped at random, and knowing which five tend to go missing tells you what to fix.

Save to Pinterest

You look at the photo and count nine garments. The results list four. Nothing broke, and the five that are missing were not chosen at random.

Five categories go missing consistently, each for a different reason, and four of the five have the same fix.

What has to happen for a garment to be read?

It has to present enough evidence to be described.

Snagfit reads the photo and returns up to eight garments, ordered by prominence, with a description of each covering the type, the colour, the material, the pattern and any visible detail. The full pipeline is in how outfit search actually works.

For a garment to make that list, it needs to occupy enough of the frame to be distinguishable, have a readable edge separating it from what is next to it, and be visible enough that a description would be a description rather than a guess.

A garment that fails any of those is not described badly. It is absent.

That distinction matters, because it changes what you do about it. A bad description means re-cropping might improve the words. An absence means the garment was not in the photo in any useful sense.

Why do small accessories disappear?

Because of the downscale, and it is arithmetic rather than judgement.

Every upload is reduced to 1280 pixels on the longest side. A full-length photo of a person is then roughly 1280 pixels of body from head to foot. A belt occupies a small band of that. A ring occupies a handful of pixels. A pair of stud earrings occupies almost none.

At that size there is nothing to describe. The pixels exist and they carry a colour and no shape.

This is why jewellery is the hardest category, covered in finding jewellery and small accessories, and why sunglasses come back generic from a wide shot, covered in finding sunglasses from a photo.

The fix is a crop, and it works as long as the original photograph had the resolution to survive it.

Why do layered garments go missing?

Because layering hides most of them and fragments the rest.

A shirt worn under a jumper shows at the collar and the cuffs. That is a small percentage of the garment, and it is not enough to say anything about the rest of it. Describing it would mean inventing the other 90 percent.

A garment worn under something and not visible at all is simply not in the picture. No crop recovers it.

Then there is fragmentation. An open coat is two vertical strips of fabric with something else between them. A bag strap divides a top into two halves. A scarf splits a jumper into an upper and lower section. A garment broken into disconnected pieces presents less evidence than the same garment whole, so it can lose its slot to something plainer beside it, which is the mechanism in why the search picks the wrong garment.

For the fragmented case, cropping to the garment usually brings it back, because removing the competition changes the ordering. For the genuinely hidden case, a different frame is the only option.

Why does contrast matter so much?

Because an edge is what defines a shape, and contrast is what makes an edge.

Black trousers against a dark floor have no boundary at the hem. A white shirt against a white wall has no boundary at the shoulder. A tan coat against a sand-coloured building has no outline at all.

Without a boundary there is no shape, and without a shape there is nothing to identify. The garment either goes missing or gets merged into whatever it is touching, which occasionally produces a description of something that does not exist.

Low light does this generally, because a dim photograph has less contrast everywhere. So does a busy background with similar tones. And so does a very bright photograph where the highlights blow out, which is the specific problem white garments have in direct sun.

The fix here is not a crop. It is a different photograph, ideally in daylight in open shade, for the reasons in why lighting changes what a search finds.

Why do shoes so often not appear?

Four things happening at once, and shoes get all of them.

They sit at the bottom edge of the frame, which is where photographs are most often cropped. They are small relative to the body. They are frequently in shadow, since light usually comes from above and feet are the furthest thing from it. And they are often partly cut off or obscured by the hem of a trouser.

The result is that shoes are the single most commonly missing item in a full-length outfit search, despite being one of the things people most want to find.

Cropping to the feet fixes it when the original has the resolution. The dedicated version of this is in finding shoes from a photo, and the short version is that a lower camera angle helps more than anything else.

Can the background take a slot?

Yes, and it is the case people find most surprising.

Anything in the frame that looks like a garment is read as one, because it is one. A coat over the back of a chair. A jumper on a bed. A rail of clothes behind the subject. A dressing gown on a door. A towel, sometimes, since it is a large rectangle of textile.

Those occupy slots and attention that the outfit needed. In a photo taken in front of an open wardrobe, most of what is in frame is clothing and only a small part of it is being worn.

Cropping the background out is the fix, and it is the same fix as most of this post. It is also the fix for a second person in the frame, which is the same problem with a person attached to the extra garments.

Does the eight-item cap ever matter?

Rarely for one person, often for more than one.

A single-person outfit seldom contains more than eight distinguishable garments. Coat, top, trouser, shoes, bag, hat, scarf, belt is already eight, and most outfits have fewer than that visible at once.

Where the cap bites is a group photo, where the eight slots are shared. Two people means about four each. Three means fewer, and the smaller items on everybody drop off entirely.

So if you are looking at a photo of one person and something is missing, the cap is almost certainly not the reason. If you are looking at a group, it may well be.

Does a garment ever get counted twice?

Occasionally, and it looks like the opposite problem.

A garment with two visually distinct parts can be read as two garments. A dress with a contrast bodice and skirt, a two-tone coat, a jumpsuit with a belt at the waist, or a top and trouser in the same fabric that read as separates.

The reverse happens too: a top and a skirt in the same colour with no visible waistband can be read as one dress, which is the case described in finding a skirt from a photo.

Both come from the same thing. Snagfit reads continuous regions of fabric and describes them, and a garment is not always one continuous region. A jumpsuit is one product and two regions. A matching set is two products and one region.

Neither is common enough to plan around, and both are fixed the same way: crop to the piece you want, so there is only one thing in the frame to read.

Should you count on the results being complete?

Treat the list as what the photograph carried rather than as an inventory of the outfit.

That framing avoids two mistakes. The first is assuming a missing garment means the search does not work, when it usually means the photo did not show it well enough. The second is assuming the list is exhaustive, and not looking at the picture yourself for the piece you actually wanted.

The most useful habit is to look at the photo first and decide what you want before you upload. Then you know whether the result answered your question, rather than reading whatever came back and taking that as the answer.

If the piece you wanted is not in the list, that tells you something specific: it was small, hidden, low contrast, fragmented, or at the edge. Each of those points at a different next step, and the first four of them point at a crop.

What is the actual fix?

Crop to the missing garment and upload again.

That single action does three things at once. It gives the garment the full 1280 pixels instead of a small share of them. It removes the competition that pushed it down the ordering. And it drops the background objects that were taking slots.

The exception is a garment that is genuinely hidden rather than small. Cropping cannot reveal what the photograph did not capture, and the only fix there is a different frame. If the source is a video, that frame usually exists.

Worth deciding before you start: each upload uses a look, so working out which two or three pieces you actually want is more efficient than uploading everything. That triage is what shopping your camera roll is built around, and it is a faster way to work than uploading the same photo four times.

Common questions

Why does an outfit search miss some of the clothes in a photo?

Because a garment has to present enough evidence to be read as a garment. Snagfit returns up to eight garments per photo, ordered by prominence, and anything small, partly hidden, low in contrast against its surroundings, or fragmented by something worn over it may fall below that threshold. The missing pieces are consistently the same kinds of pieces rather than a random selection.

How many garments does Snagfit return per photo?

Up to eight, ordered by prominence, which tracks roughly how much of the frame each garment occupies and how clearly it reads. A single-person outfit rarely exceeds that limit, so the cap is seldom the reason something is missing. In a group photo the eight slots are shared across everyone in frame, and that is when the cap starts to matter.

Which garments most often go missing from a search?

Small accessories, anything worn under something else, garments that match their background in colour, and pieces broken into fragments by layering. Socks, belts, jewellery, a shirt under a jumper, black trousers against a dark floor, and a top divided by a bag strap are the recurring cases. Each has a different cause and they respond to different fixes.

Does a missing item mean the search failed?

Not usually. It means the garment did not present enough evidence in that photograph to be described, which is a fact about the picture rather than a fault in the search. A garment that is 90 percent covered is not really in the photo in any useful sense, and describing it would mean guessing. Snagfit describes what it can see.

How do you get a missing garment into the results?

Crop to it and upload again. That gives the garment the full frame and the full resolution instead of a small share of both, and it removes the competition that pushed it down the list. If the garment is genuinely hidden rather than small, a different frame from the same source is the only fix, since cropping cannot reveal what the photo did not capture.

Why do shoes often not appear in a full-length photo?

Because they sit at the bottom edge of the frame, they are small relative to the body, they are often partly cut off, and they are frequently in shadow. Snagfit downscales every upload to 1280 pixels on the longest side, so a pair of shoes in a full-length shot gets a small share of those pixels. Cropping to the feet is what brings them back.

Can a background object take a slot from a real garment?

Yes. Anything in the frame that looks like a garment is read as one, including a coat on a chair, a rail of clothes, a towel or a garment on a hanger. Those occupy slots and attention that the outfit needed. Cropping the background out is the fix, and it is the same fix as most of the other cases.

Is it better to upload once or several times?

Once for an overview and then separately for the pieces you want. A single upload of the whole outfit tells you what is there. A cropped upload per garment describes each one properly. Each upload uses a look, so it is worth deciding which two or three pieces you actually care about before you start rather than uploading everything.

Keep reading

How it works

How do you find an outfit from an iPhone screenshot?

An iPhone screenshot is capped at your screen's resolution, not the video's, and it saves as HEIC by default. Both of those change what Snagfit, or any tool, can actually read.

6 min read
How it works

What does an AI see in an outfit photo?

Five fields per garment, eleven possible categories, and a search query four to eight words long. Here is the exact shape of what comes back, with a real example.

7 min read
How it works

Can you find clothes from a screenshot?

Yes, usually. Whether you find the exact piece or something close depends on three things about the screenshot, and you can check all three before you upload it.

7 min read

← Back to Snagfit