How Omniris finds objects in your drone photos and maps with AI, and how a person checks and corrects every result before anyone relies on it.
Drone flights produce thousands of photos and maps far larger than anyone can study inch by inch. Omniris uses AI to find the things you care about in them: vehicles, buildings, solar panels, machinery, debris, livestock, or anything else you can name in a few words. Each object found comes back outlined, boxed, labelled and counted.
AI is fast, but it is not always right. So Omniris treats every AI result as a first draft. A person can check it, correct it and mark it Verified. The check is recorded, with who made it, when, and exactly what they changed. The AI's own answer is never overwritten, so both statements survive: what the AI said, and what a person confirmed.
Type what to look for, or pick a ready-made Category. No training, no labelled examples, no data science.
Grounded SAM 2, OWLv2 + SAM 2.1 and SAM 3, each with its own strengths, all in the same dialog.
Single photos, whole folders, or orthomosaics many gigabytes in size, read piece by piece at full detail.
Move, resize, relabel, delete or add boxes and outlines, then tick Verified. Work through a folder with Verify & next.
The same editor everywhere, including on the phone with no signal. Offline checks are sent when the phone reconnects.
Every verification is sealed in the image's tamper-evident custody record. The AI's original answer stays one click away.
Detections become findings, feed change between two dates, and export to GeoJSON, CSV and KML.
Object detection, the kind of AI that finds and outlines things in a picture, has become remarkably capable. It still makes two kinds of mistake. It misses things that are there, and it marks things that are not. Seen from above, a shadow can pass for a vehicle, a skylight for a solar panel, a tarp for a pool.
For a count on a dashboard, a few mistakes may not matter. For an insurance claim, a compliance report, a dispute or a work order, they do. The usual answers both fall short:
Omniris does both inside one record. The AI does the first pass. A person checks the result in an editor built for it. Their corrections become a separate, verified version, and the image's custody history records the check. From then on the verified version is what everyone sees: in the image viewer, on shared links, in exports and in comparisons between dates.
All three models are open-vocabulary: they are not limited to a fixed list of objects. You describe what to find in words, and the model looks for it. Each model also traces an outline around every object it finds, not just a box, so you can see exactly which pixels it means. You pick the model at the top of the AI Detection dialog; Grounded SAM 2 is selected unless you choose another.
| Model | Good at | Keep in mind |
|---|---|---|
| Grounded SAM 2 detection + outlines | A general-purpose detector. One pass reads the whole prompt, so a longer prompt costs nothing extra. Smaller tile sizes help it find small objects in large images. | A long list of classes makes it start mislabelling objects rather than simply missing them, so name only what you expect to see. No Category presets. The dialog suggests SAM 3 for aerial subjects where it is available. |
| OWLv2 + SAM 2.1 detection + outlines | Each class you name is a separate question, all answered in one pass, so prompt length is free. Its licence carries no restrictions on how the results are used, which suits customers who cannot accept use restrictions. | Its confidence scores run lower than the other models', so its default box threshold is lower (0.2 instead of 0.45). The text threshold does not apply to it. One tile size only. |
| SAM 3 concept segmentation | Finds and outlines every instance of a short concept such as "solar panel" or "excavator". The dialog describes it as stronger on unusual aerial objects. Adding nouns does not make it worse at the others. Ready-made Categories. | It looks for each concept separately, so every extra noun adds to the work and can add to the price. Its licence (Meta's SAM License) asks you to acknowledge SAM if you publish results, and prohibits military and unlawful use; the dialog shows this notice before you start. |
You can run more than one model on the same images. Each image keeps one result per model, side by side, and the image viewer lets you switch between them. Running a model again on images that already have that model's results replaces them; the dialog warns you first, and says which other models' results will be kept.
The prompt is the list of things to look for, typed into Objects to detect. Separate the classes with full stops (commas work too):
The default prompt. Your last prompt is remembered in your browser for next time.
Capitals and repeats are ignored, so "Car. car." is one class. The classes you name become the labels on the results, and later the list a person picks from when correcting them.
For SAM 3 and for OWLv2 + SAM 2.1, a Category menu above the prompt offers ready-made prompts for common kinds of work. Picking one replaces the prompt with that Category's list; you can still edit it, and the menu then reads Custom. Each Category is deliberately short, because on SAM 3 every noun adds work. Where a Category has limits you should know about, a note appears under the prompt when you choose it.
| Category | Looks for | The note shown with it |
|---|---|---|
| SAM 3 | ||
| General | building, person, car, truck, van, bus, tree | None. The same list as the default prompt. |
| Solar PV | solar panel | Finds panels, not defects: cracks, soiling and hot spots are not what it looks for. |
| Roof & property features | roof, tarp, skylight, debris | A detection aid only: it finds roof features, not damage. Whole roofs wider than about a quarter of the frame split across tiles; raise the tile size. |
| Construction site | excavator, dump truck, wheel loader, tower crane, concrete mixer truck | Finds common site machinery. Expect to tune the box threshold, and check against a known image before relying on the counts. |
| Powerline corridor | transmission tower, power line, tree, building | Looks for corridor features, not tower hardware: insulators, spacers and dampers are not looked for. A power line is a thin, linear shape, the hardest kind for this model, so expect broken-up conductors. |
| Storm & flood response | flooded building, flooded road, water, debris | Condition words like "flooded" are the weakest prompt shape for this model. Check against a known-good image before relying on the counts. |
| Livestock & pasture | cattle, sheep | Works best on open pasture: animals under trees or inside sheds are hard to see from above. Check against a known image before relying on the counts. |
| OWLv2 + SAM 2.1 | ||
| Site monitoring | car, truck, bus, building, shipping container, excavator | A detection aid, not a census: it misses some vehicles and buildings and marks some things that are neither. Check against a known image before relying on counts. |
Grounded SAM 2 has no Category menu. Nothing is blocked: you can type any of these words into its prompt yourself.
Start a job from wherever the images are:
The dialog opens at once and fills in the image count and price as it works them out. Nothing is charged until you press Start.
flowchart LR A[Photos, folders
or orthomosaics] --> B[Model, prompt,
tile size, thresholds] B --> C[Price shown
before you start] C --> D[Each image cut into
overlapping tiles] D --> E[Outlines, boxes,
labels, confidences] E --> F[A person checks
and verifies] F --> G[Findings, change,
exports, reports]
A detector looks at a picture of a fixed size. A large photo or map is therefore cut into overlapping squares, called tiles, and each tile is examined at full detail. This is what lets a small object in a big image be found. By default neighbouring tiles overlap by a fifth.
| Model | Tile sizes offered | What the choice does |
|---|---|---|
| Grounded SAM 2 | 800 · 600 · 400 · 200 px | 800 px is the standard price. Smaller tiles find smaller objects but multiply the work, and the price (see Price). |
| OWLv2 + SAM 2.1 | 960 px | The model's own size, so tiles reach it without rescaling. At the standard price. |
| SAM 3 | 1008 · 1512 · 2016 px | 1008 px is the model's own size and gives the best detail. Larger tiles cost less (fewer tiles per image) but shrink small objects. Pick one when your subjects are wider than a tile, such as roofs, roads, fields or water, and you want each captured as one object. |
Images whose results a person has verified are skipped by default, and not charged for. The dialog says how many were skipped. Tick Include verified images to run them again: their verified version stays what everyone sees, the new AI result is kept beside it, and the image is marked Changed since verification so someone can review it. Only people who could verify in that folder can tick it.
The job appears on the AI tab with a progress bar and a plain status, such as "Queued — waiting for GPU". You can cancel it there, view or download its log, and retry a failed job with the same model, prompt and settings. The job carries on if you close the page, and you get an email when it finishes or fails. Results then appear on each image.
Every object the AI finds comes back with four things:
In the gallery, each image's card shows how many objects were found, with a Verified badge once a person has checked it. Open an image and the AI Detection panel shows the rest:
| In the panel | What it does |
|---|---|
| Counts by class | One tile per label with its count. Click a tile to show or hide that class on the image. |
| Confidence slider | Hides detections below the confidence you set, on the image and in the list, and updates the counts. Boxes a person added have no confidence and are always shown. |
| Detection list | Every detection with its label and confidence. Click one to show only that object. |
| Model picker | Appears when the image has results from more than one model. |
| Mask Overlay | Shows or hides the outlines as a coloured layer. |
| Verification | Whether the result is verified, by whom and when, and the buttons to edit and verify (next section). |
| Export detections | This image's detections as GeoJSON, CSV or KML. |
| Findings | Promote visible turns the detections now shown into findings; a flag on each row turns one detection into a finding. |
| Remove Detections | Deletes the AI result from this image. The image itself is untouched. |
On a large orthomosaic the outlines are fetched in pieces as you pan and zoom, so the whole map's outlines never have to load at once.
The job card on the AI tab gives the same results for the whole job: Open the folder, Export detections for every image in the job, and Findings for what the job's automatic findings created.
Orion, the Omniris AI assistant, can answer questions about the results in plain language: which images contain a class, how many objects were found, how many images are verified, or what one model found that another missed. See Orion, the AI assistant.
Verification means a person has looked at an image's AI result, corrected what was wrong, and confirmed it. It happens in the image viewer, in the same panel as the results.
Anyone who can edit the image's folder, including a client working on results a pilot delivered to them, in a vault they can edit. People with view-only access see the verified version and who verified it, but cannot change it. Editing and verifying are free.
Press Edit. The image becomes yours to edit for as long as you work on it, and anyone else who opens it sees that you are editing it. If you left it open on another device, you are asked whether to continue here.
| To | Do this |
|---|---|
| Remove a false alarm | Select it and press Delete |
| Fix a box | Drag it to move it, or drag a corner to resize it |
| Fix a label | Pick the right one from the label list |
| Add a missed object | Add box, then draw it |
| Fix an outline | Outline: move, add or delete its points, redraw it, or remove it |
| Step through objects | Tab and Shift+Tab |
| Take something back | Undo and Redo (Ctrl+Z, Ctrl+Shift+Z) |
The label list is the words the AI job was asked to look for, plus any class the AI returned. People correct results in the job's own vocabulary, so a "truck" never becomes a "lorry" on the next image.
The first time anyone edits an image, its AI outlines are turned into editable shapes ("Preparing outlines…"). Very detailed outlines are simplified to at most 5,000 points. On a large orthomosaic this is done in the background; if it cannot be done, you can still correct the boxes.
Each verification is written as a sealed event in the image's tamper-evident custody record, the same record that tracks its upload, processing and sharing. The event records:
The panel's History lists every verification with its change summary, and links to the custody record. Custody certificates for the image list the AI result and each verification. The companion paper Evidence you can prove explains the custody record and certificates.
What a verification records, in summary. Illustrative example values.
A verification stays in force until something real changes underneath it:
When a detection you change is the evidence for a finding, the editor names the finding and asks what to do: Keep the finding, or mark it a False positive. With Keep, the finding follows your correction: its evidence picture is re-cut from the corrected box and it is moved to the corrected place, unless someone set its location by hand. A deleted detection stays in the finding's evidence, marked as removed in that verification. Each decision is applied on its own, so one that cannot be applied (for example, a finding under legal hold) is skipped and listed, and the verification still goes through.
Checking is often best done close to the site, by the person who flew it. The Omniris phone app and desktop application use the same editor as the web, so the tools, the rules and the record are identical wherever the check is made.
The phone's editor is built into the app, so it opens without a connection. To prepare a folder for a site with no coverage, choose Make available offline for verifying on the folder while you still have a connection. The app downloads each image and its AI results, with a permission to verify it offline that lasts 30 days.
On a connection, make the folder available offline. The app reports what it stored, what is still being prepared, and anything it could not take, with the reason.
Correct and verify as usual, in airplane mode if need be. Your edits are kept on the phone.
Each offline verification waits on the phone, with Retry, Review and Discard, until there is a connection.
It is accepted and sealed as long as nobody else verified the image in the meantime. If someone did, you are asked to review the two versions.
Some images can only be checked on the phone while online:
Signing out deletes the downloaded images and results from the phone. Your unsent drafts are kept, hidden from anyone else, and resume when you sign back in. The companion paper the phone app in the field covers the app as a whole.
When the same site has been mapped on two dates, Omniris can compare what the AI found on each and tell you, object by object, what appeared and what disappeared. It works in the Compare view of the Site Navigator: pick the two dates and a map layer, choose the model, set the confidence, and press Start.
Both dates must be orthomosaics (map-accurate GeoTIFFs) with results from the same model and the same prompt. If the prompts differ, the comparison is refused, so that a class nobody looked for on one date is never reported as disappeared.
Each date uses its verified result where a person has verified it, and the AI's otherwise. Detections below the confidence you set are left out on both dates; it starts at the model's own default.
Every detection is placed on the ground with a stated margin of error. Two maps at very different ground resolutions (more than twice apart) are refused, because a sharper map always shows more, which would read as false "appeared" objects.
Two maps of the same place are rarely aligned perfectly. The shift between them is estimated from buildings and other fixed structures and allowed for. If it cannot be estimated reliably, the result says so, and why.
An object of the same class found on both dates, within the combined margins of error, is unchanged. One found only on the later date appeared; only on the earlier, disappeared.
The results are drawn on the map: appeared in green, disappeared in red, and unchanged in grey if you choose to show them. Click an object to see its confidence on each date, how far it moved and its margin of error. The summary gives the counts and the shift between the dates. Distances are in feet, with metres in brackets.
Each comparison is recorded as an event in the later image's custody record. Objects that appeared or disappeared are filed as findings awaiting review, and the comparison can be printed in the situation report (see Reports and exports). A comparison reuses stored results, so it is not charged for.
A detection says "there is a tarp here". A finding is the issue someone has to act on: placed on the map, with a category, a severity, evidence pictures, an owner and a status. Omniris turns detections into findings in several ways:
| Where | What happens |
|---|---|
| When you start the job | One choice of three. No automatic findings. When the job finishes: the job's detections become findings in one run. When each image is verified: as a person verifies an image, its verified detections become findings, and nothing is created for an image until it is verified. |
| Per class, at the start | For each class in the prompt, choose the finding's category and severity, or leave it to your organization's mapping rules. |
| Later, on a folder | Create findings from AI results: from the detections already stored, limited to the classes and minimum confidence you choose, with or without subfolders. |
| On one image | Promote visible turns the detections now shown (after the class and confidence filters) into findings; the flag on a single detection turns just that one. |
| From a change comparison | Objects that appeared or disappeared between two dates are filed as findings awaiting review. |
Detections close to each other are grouped into one finding, and ones near an open finding are added to it rather than duplicating it. The choice When the job finishes is remembered for the organization or vault; When each image is verified must be chosen each time.
When the job finishes files the AI's detections straight away, mistakes included, for people to triage. When each image is verified files only what a person has confirmed. The second is slower to fill up, but nothing in it is unchecked.
The companion paper Findings you can act on describes findings in full: triage, assignment, evidence, history and reports.
Export detections is on an image's AI panel, on a folder (with every folder below it), and on an AI job's card. It writes the detections as GeoJSON, CSV or KML, for mapping (GIS) software, spreadsheets and Google Earth.
Generated from the Compare view for a site and two dates, the situation report is one PDF that brings the comparison together:
It is written in English or Polish, and saved as a new file in the later map's folder (or your own Reports folder) with its own custody record. Encrypted imagery is never printed into it.
For sharing folders, links and delivery to clients in general, see sharing, delivery and integrations.
AI detection is priced by the number of images, in blocks of up to 100. The standard rate is 14 credits per block, rounded up to whole blocks: 1 to 100 images cost 14 credits, 101 to 200 cost 28, and so on. The dialog shows the price, and your credit balance, before you start.
| Model | How the price is set |
|---|---|
| Grounded SAM 2 | The standard rate at 800 px tiles. Smaller tiles multiply it: 600 px ×2, 400 px ×3.5, 200 px ×11, because they multiply the work per image. The prompt's length does not affect the price. |
| OWLv2 + SAM 2.1 | The standard rate. The prompt's length does not affect the price. |
| SAM 3 | The standard rate covers a typical drone photo (about 5 megapixels) looked at for up to seven concepts at 1008 px tiles. A job that does more work per image, because the images are larger (more tiles) or the prompt longer (more concepts), costs proportionally more. A job that does less never costs less than the standard rate. Each tile size is priced separately in the dialog. |
Detection runs on a processing machine with a graphics processor (GPU), not in your browser. Once started, the job is carried through by Omniris itself, so you can close the page.
| Deployment | Detection runs on |
|---|---|
| Omniris cloud (Microsoft Azure) | GPU processing machines in the Omniris cloud |
| Omniris cloud (Google Cloud) | GPU processing machines in the Omniris cloud |
| Data center edition and on-premises | The site's own GPU workers; no images leave the site unless you choose Hybrid for a job (below). SAM 3 needs a GPU with at least 8 GB of memory. If no machine has one, the dialog says so, and a SAM 3 job is refused before anything is charged; the other models still run. |
A site running Omniris on its own hardware can also connect to the Omniris cloud. The dialog then offers Run in Omniris Cloud (Hybrid), job by job: the images for that job are processed in the Omniris cloud and the results come back to the site. It is never chosen for you.
Editing and verifying need no GPU and work in every deployment, on the web, the desktop application and the phone.
AI detection is a fast, consistent first pass. These are its limits, stated plainly.
An AI result tells you where to look. A verified result tells you that a named person looked, when, and what they changed. The situation report puts it plainly: AI detections can be missed or wrong, and items not reviewed by a person have not been checked, so confirm them on the imagery before acting on them.
Find tarps, skylights and debris across a neighbourhood of roofs, have an adjuster verify each image, and hand over a record showing who checked what.
Count and outline panels across a site, verify the count, and export it to your mapping software. Pair it with the thermal survey for hot spots (Temperatures you can measure).
Track excavators, cranes and trucks across flights, and compare two dates of a site to see what arrived and what left.
Locate towers, lines, trees and buildings along a corridor. For measured clearance to the line, see LiDAR you can work with.
Flag flooded buildings and roads and debris after an event, verify in the field without signal, and issue a situation report for the two dates.
Count cattle and sheep on open pasture, flight after flight, with every count checked before it is reported.
Vehicles, buildings, containers and machinery, compared between dates, with what appeared and disappeared filed as findings for review.