How to Use Scout: A Complete Tutorial

A practical guide to inference, orthophotos and training your own detector

Open the portal at TreeTools Scout.

TreeTools Scout helps you detect and count objects in aerial imagery, review detections over an orthophoto, download GIS results, and train a detector using your own examples. This tutorial follows the current user interface. Button names appear in bold so you can match each instruction to the screen.

You do not need to install desktop processing software to use the portal. You need a TreeTools website account, suitable imagery, an internet connection and, for a new chargeable inference project, a positive available credit balance. Sign in through WordPress on treetools.ai; the Scout workspace uses that account automatically. Keep your own copies of important source imagery and download any results you need before they expire.

1 Choose the right workflow

The two main work areas are Inference and Detector. Inference applies an existing model to imagery. Detector is where you prepare examples and train a model of your own.

An orthophoto is a single image assembled and corrected so that locations correspond to positions on the ground. A folder of raw flight images is not an orthophoto yet. Choose the source that matches the files you actually have: an existing TIFF uses Inference; a raw-image folder uses Ortho & inference.

A boundary patch limits inference to the area you want to assess. It is different from a training AOI. The boundary belongs to an inference project; training AOIs divide a detector's training image into smaller example areas.

If you already have a suitable model, start with inference. Train your own detector when you need to supply examples of the objects and imagery conditions you want it to learn.

2 Sign in and find your projects

1. Open TreeTools Scout in a current browser on a computer with a stable internet connection.
2. If you see a login message, use the website's Login link and sign in to your TreeTools account. New users can use Register on the TreeTools website to create an account.
3. Return to the Scout product page. If your dashboard shows a Scout card, Open Scout opens the same portal.
4. Wait for the embedded Scout workspace to open. You do not need to enter a second Scout username or password.
5. Check the name beside Hi, in the workspace, then use Inference or Detector. Confirm you are using the intended account before uploading imagery or starting a run.

Your projects and trained detectors are associated with the signed-in account. Another account's projects do not become available simply because both accounts use the same portal.

The credit balance is displayed with an ha label and one decimal place. Click it to refresh the balance. New accounts start with 0 credits. Use Add credits to purchase a pack before starting chargeable inference. Read the pricing explanation in section 4 before processing, particularly if you plan to use Nursery. A displayed balance of 0.0 can be a rounded value; the server checks the underlying balance when you submit work.

Close workspace closes your Scout session. Use the TreeTools website's Logout option to sign out of WordPress as well, especially on a shared computer.

On the inference page, All shows your project list. Use the star beside a project to mark it, then use Starred to return to marked projects. A star helps you organise the list; it does not extend storage retention.

For large uploads, connect the laptop to power and prevent it from sleeping. Keep the browser tab open until the upload has finished and the project has been submitted. You can move between pages within TreeTools Scout while an inference upload continues. Avoid refreshing, closing the tab or signing out during that transfer.

If your session expires, sign in again. Before repeating a submission, check whether the project already appears in your list. This helps avoid creating a second project for an upload that was already accepted.

3 Prepare your imagery and boundary files

Check the limits before starting

GiB and MiB are binary file-size units. A 20 GiB limit is approximately 21.47 GB in decimal units. Use the file properties on your computer to check size; the portal also validates uploads.

Prepare an existing orthophoto

Choose the original georeferenced TIFF rather than a screenshot, PDF or JPEG export of the map. The image needs valid ground coordinates for boundary alignment and area measurement. If you are unsure whether it is georeferenced, ask the person who created it or check it in your GIS software.

Use a recognisable filename and a project name that includes the site, block or survey date. The project name can be changed in the creation form without renaming the local file. For example, North Block — October 2026 is easier to identify later than image_final_2.

Prepare folders of raw images

You can combine several raw-image folders in one inference project. Click Add image folder for each folder you want to include, or select a parent folder to include its subfolders in one step. Each addition keeps the folders already selected. Scout selects JPG/JPEG images and skips other file types.

Check the folder list and combined image count before processing. Use Remove to exclude a folder or Clear all to start the selection again. The combined selection must contain no more than 20,000 images and 20 GiB in total; this limit applies to the whole project, rather than separately to each folder.

All selected images are uploaded into one flat image set for orthophoto generation. Scout adds unique numbered prefixes to the upload filenames, so Flight1/IMG_001.JPG and Flight2/IMG_001.JPG can both be included. Your local files and the images' contents, including camera metadata, remain unchanged. Keep an unchanged backup of the originals.

Raw imagery must be suitable for reconstructing an orthophoto. Check that the images belong to the intended survey and have the coverage needed for that reconstruction. A raw folder's compressed size does not predict the final TIFF size: the generated orthophoto can be larger, and the 20 GiB preview limit still applies to it.

Prepare a boundary patch

For a shapefile, keep the matching files together. At minimum include .shp, .shx and .dbf. Include .prj so the coordinate reference system is declared. Other matching companion files, such as .cpg, may also be included. A ZIP containing one complete shapefile set is a convenient way to share the boundary without losing a component.

Use one intended boundary geometry. Check that it covers the assessment area and overlaps the orthophoto in the correct ground location. A boundary drawn in the wrong location can produce missing detections or an invalid-patch error.

The coordinate reference system, or CRS, tells the application what the coordinates mean. A shapefile's .prj is therefore important. In this workflow a GeoJSON without an explicit usable CRS declaration is treated as having no CRS; do not rely on a GIS program's assumed default. If in doubt, provide a shapefile set with its .prj.

Boundary files are checked when processing starts. Patch pending after upload is normal. If a boundary has no usable CRS, it is not applied and inference continues on the full TIFF. The project can show Patch not applied with an explanation. Unreadable or invalid geometry can stop processing and must be corrected.

For Ortho & inference, a generated orthophoto is converted to a valid uploaded boundary's CRS when the systems differ. If there is no boundary or no usable boundary CRS, the orthophoto is left unchanged. There is no automatic requirement for every project to use EPSG:2193.

4 Understand credits before processing

Buy and top up credits

WordPress holds your Scout credit balance, and MemberPress handles checkout. Every new account starts with 0 credits. Choose Scout 1,000 ha, Scout 2,000 ha or Scout 2,500 ha using Scout Plans or Add credits in the Scout workspace.

These are one-time credit packs. Each successful purchase adds its hectares to your remaining balance; it does not replace the existing balance or start an automatic renewal. You can buy the same pack again or choose a different pack. A purchase also adds to a negative balance: for example, a 1,000 ha pack added to a balance of −20 leaves 980 credits.

After payment, return to Scout and click the credit balance to refresh it. If a completed purchase has not appeared, contact TreeTools support with the information shown in your MemberPress account. There is no need to buy the same pack again while asking about an unconfirmed payment.

Scout warns when your available credits are below 100 ha. This uses the balance after reservations for work already in progress. Add credits before starting a new chargeable project if the available balance is zero or negative. You can still open the workspace and access projects when the balance is not positive.

How inference uses credits

Credits are based on hectares, with a model-specific rate. For standard inference, one hectare uses one credit. Nursery uses 900 credits per hectare. The balance may still be displayed with a ha label; when using Nursery, the same numerical balance is consumed at the 900-times rate.

The first successful inference on a project charges the usable boundary area when available. If no boundary can be applied, it uses the TIFF area. This is why a carefully prepared boundary can prevent paying for irrelevant ground around the area you need.

Each inference project records its largest paid area. A later run is charged only for growth beyond that maximum. Changing the selected model alone does not incur another area charge, including a switch between standard and Nursery models. The rate for any newly added area is the selected model's rate for that run.

The last two rows are alternative examples. Once the project has paid for 12 ha, a later model change at 12 ha is free. Reducing the area does not undo the earlier charge; the project retains its largest paid area.

A new project or an increase in paid area requires a positive balance when admitted. The application permits that run to leave a negative balance. For example, an account with 5 credits can start a standard 10 ha project and finish at −5 credits. It will then need additional credits for a new project or further area growth.

An existing project's model-only rerun within its paid area is allowed at zero or a negative balance. If you replace its boundary, the application checks the actual area before permitting additional unpaid area to proceed.

The Paid area displayed with a project is the paid maximum in hectares. It is not a union map of all earlier boundary shapes. Charges are tracked per project: creating another project with the same imagery does not carry the first project's paid area across.

Credits can be reserved while work is running; the charge is committed after successful inference. A confirmed failed or cancelled job releases its reservation. If processing fails, read the error and correct the cause before retrying. Use Add credits to top up, or contact TreeTools support to clarify a balance, then click the balance in the header to refresh it.

5 Run inference on an existing TIFF

1. Open Inference and click + New project.
2. In Create inference project, enter a Project name. Use a site or survey name you will recognise later; the limit is 100 characters.
3. Set Imagery source to Orthophoto TIFF.
4. Click Choose TIFF and select one .tif or .tiff. Check that the selected filename is correct.
5. Under Patch / boundary files, click Choose patch files and select your ZIP, GeoJSON or complete shapefile set. Confirm the filenames shown beneath the button.
6. Choose the Detection AI model.
7. Check Inference type. With an existing TIFF it is Inference.
8. Review the credit information, then click Process once.
9. If no boundary is selected, read Process without a boundary patch?. Choose Add boundary to return to the files, or Continue without boundary if you deliberately want the whole image processed.
10. Watch the upload card above the project list. Keep the browser tab open until the transfer and submission finish.

The form may suggest a project name from the TIFF filename. You can replace that suggestion before clicking Process. Choosing the file does not run inference immediately; the Process action records your complete settings and starts the upload.

Current built-in model names are Age 3-4 V3 (Built-in), Post Thinning V8 (Built-in), Nursery V1 (Built-in) and Cutover V2 (Built-in). Your account's available trained detectors can also appear. Choose the model appropriate to the imagery and objects you want to detect; review results on a representative area before relying on a large assessment.

If the model list is still loading, wait for it to finish. If it reports an error, use Retry loading models. An unavailable model cannot be used simply by retaining an old selection.

Once the upload is verified, the project enters the processing queue automatically. You do not need to wait for the upload to finish and then click Inference a second time. Check the project's status to confirm whether it is queued, running or completed.

6 Generate an orthophoto and run inference

1. Open Inference → + New project.
2. Enter the Project name.
3. Set Imagery source to Folder of raw images.
4. Click Add image folder and select a survey folder. Repeat to add more folders from the same survey, or select a parent folder containing them.
5. Review the selected folders and combined JPG/JPEG image count. Remove unwanted folders and check the total size. These images will be combined into one orthophoto, so include only images belonging to the intended assessment.
6. Add the optional boundary under Patch / boundary files. Use a boundary with a declared CRS.
7. Select the Detection AI model, taking account of the Nursery rate if applicable.
8. Confirm Inference type is Ortho & inference. Raw images use this mode automatically.
9. Click Process, respond to any no-boundary warning, and leave the browser tab open while the files upload.
10. Wait for submission, orthophoto generation, inference and final publication to complete.

This workflow does more work than inference on an existing TIFF. It first reconstructs an orthophoto and then applies the selected model. A large number of photographs can take substantial time to transfer and process. Progress may move differently through the orthophoto and detection phases; it is not a reliable countdown.

View stays unavailable until an orthophoto exists. When it becomes available, you can request the preview. If another operation on the same project is unfinished, preview preparation may wait for it.

After a verified orthophoto is stored, the original raw images are removed from the project's application storage. Keep your original survey photos in your own archive. The retained orthophoto allows a later inference retry to use the generated TIFF, including when detection fails after orthophoto generation.

Use Download ortho to obtain the retained TIFF when available. Generated orthophotos include the project name in the filename: for example, a project called North Block produces North_Block_orthophoto.tif. Spaces and characters that are unsuitable for filenames are replaced with underscores. An existing TIFF that you upload keeps its original filename.

If the generated orthophoto exceeds the preview size limit, its successful generation does not guarantee that the interactive map can be prepared. Ask your administrator about a suitable smaller image or processing arrangement rather than repeatedly uploading the same folder.

7 Read uploads queues and progress correctly

There are three separate progress displays: transferring files from your computer, processing the job, and preparing map tiles for interactive viewing. Finishing one does not necessarily finish the others.

Processing capacity is shared. If another user has a job running, your project can be added to the queue. Accounts take turns, with eligible work within an account processed in submission order. The queue may explain that it is waiting for another job, processing to start, a worker, or a retry.

The same project cannot start a second unfinished operation. If you see an unfinished-job message, inspect its current status instead of clicking repeatedly. A preview request can also occupy processing time.

The inference project list shows In Progress and its percentage. It does not display processing-stage or tile-count details. The percentage covers work such as imagery preparation, detection and saving results; it can pause at a value while a long step finishes. The upload progress and map-preview progress are separate displays.

Training has its own epoch and early-stopping information, explained in section 10. Neither inference nor training percentages guarantee a completion time. A large first preview can also take much longer than a small one.

8 View the map and download results

Open the inference viewer

1. Find the project in Inference projects.
2. Wait until View is enabled. For raw images this requires the orthophoto to exist.
3. Click View to open Inference Viewer.
4. On the first visit, wait while preview preparation is queued or running. The Tiles panel shows status, progress and available detail.
5. Use the pan tool to move around, Zoom in and Zoom out to inspect different scales, and Reset view to return to the initial view.
6. Review the Summary and switch layers in Layer controls.

The first View generates display tiles when needed. Opening the project list alone does not start that preview pipeline. Once the tiles have been generated, later visits reuse them. Preview preparation and inference are separate: completed inference can still have a first preview to prepare.

Use Show boxes to inspect detection extents and Show centroids to see object centres. Show patch boundary is available when a boundary was applied. Show polygons displays polygon results where those results exist; toggling a layer does not generate missing geometry.

The summary reports Total detections, or Total detections (within patch) when the applied boundary limits the assessment. Look at representative dense, sparse, shaded and edge areas. Model output needs review; a displayed total is not a guarantee that every object has been detected correctly.

Download the deliverables

Return to the project list and open its three-dot menu:

• Download ortho downloads the retained source or generated orthophoto TIFF when available.
• Download output downloads the completed inference output ZIP, named with the project name; for example, North_Block_output.zip.
• Delete starts project deletion; use it only after saving anything you need.

The output ZIP contains final deliverables such as merged boxes and centroids shapefile sets and summaries. Applied-boundary outputs are included where generated. Intermediate per-tile TIFFs and per-tile detection outputs are not part of the user download.

The project name in the download filename helps distinguish results from different sites. Existing projects remain downloadable; an orthophoto generated before this naming update may keep its earlier filename. Save earlier ZIPs before rerunning a project if you want to compare results, because the current download contains the latest published output.

Extract the ZIP before opening a shapefile in GIS. Keep its .shp, .shx, .dbf, .prj and other companion files together. Copying only the .shp can lose attributes or coordinate information. Download the ortho separately if you also want a local raster for GIS review.

If viewing fails, Retry loading preview reloads the viewer. If tile generation itself failed, use Retry tiles after checking the error. Repeatedly creating a new inference project is not the normal way to recover a preview-loading problem.

9 Change models or boundaries and rerun

Use an existing inference project when comparing models on the same imagery. This retains its paid-area history and avoids uploading the same TIFF again.

1. Wait for the project's current inference or preview operation to finish.
2. Change Detector selected to the model you want to try.
3. To change the assessment area, use Replace patch or Upload patch in the project's row.
4. Confirm the new boundary filename and expect Patch pending until processing checks it.
5. Click Inference, Ortho/Inference, or Retry, depending on the project's state.
6. Read any no-boundary warning and wait for the new run to finish.

The list also shows Last used detector and Inferred on. These help distinguish a newly selected model from the one that produced the current completed result. Selecting another model does not immediately replace the results; processing must succeed first.

A completed project's run button can be disabled when the model and boundary have not changed. That prevents an identical duplicate run. Choose a different model or upload a replacement boundary when you intend to make a change.

Within the largest paid area, model-only reruns do not charge again. An expanded boundary can add a charge; shrinking it does not refund previous area. Check the credit rules before expanding an area using Nursery.

Download any earlier result you want to compare before rerunning. The portal's current view and download refer to the project's latest published outputs; do not rely on it as an interface for selecting historical result versions.

10 Train your own detector

Create a detector and upload the training TIFF

1. Open Detector and click Create detector.
2. In Create Your Own Detector, enter Detector Name.
3. Leave Detector Type as General, the currently offered choice.
4. Add an optional Description explaining the site, imagery or target objects.
5. Click Create.
6. In Upload Imagery, select one training .tif or .tiff and click Upload TIFF.
7. Keep the browser tab open until transfer and verification finish and the upload dialog closes. Then wait for the separate imagery preparation job.
8. Begin drawing when the map is available and Training Image reports Imagery ready.

If the upload dialog closes and the map is still preparing, that does not mean the TIFF must be uploaded again. The worker is making display tiles. If that step fails, read the message and use Retry tiling where appropriate.

Plan your AOIs

Open Preparing training data on the detector page. Follow its three principles: draw AOIs of similar size; aim for object widths around 1/10 to 1/5 of AOI width; and annotate every object fully or partly inside each AOI.

For example, a 50 m wide AOI suits objects roughly 5–10 m wide. This illustrates the relative scale, not a requirement that every AOI measure 50 m. The drawing aid recommends 640 × 640 px; use that alongside the object-to-AOI proportion and consistent sizes.

Prepare at least 8 training AOIs, 2 validation AOIs and 1 test AOI. Training areas provide learning examples, validation areas help check the model during training, and test areas support final evaluation. Choose representative areas across the image, including meaningful variation in conditions. Keep the sets distinct so that evaluation is useful.

Draw rectangular AOIs with two clicks

1. Choose Draw training AOI from the left toolbar.
2. Click one corner of the intended rectangle.
3. Move the pointer to the opposite corner. Check the displayed width and height.
4. Click the opposite corner to finish. This is a two-click action, not a click-and-drag gesture.
5. Wait for the area to save and appear in AOI Areas.
6. Repeat until you have at least eight training areas.
7. Use Draw validation AOI for at least two validation areas.
8. Use Draw test AOI for at least one test area.

Press Escape to cancel an unfinished drawing. Use the cursor tool when you want to select rather than draw. Select a saved AOI on the map or in AOI Areas before adding annotations to it.

Create a label and annotate the objects

1. Expand Label List and click + Add.
2. In Create a label, enter Label name, choose a Color, and click Create.
3. Select the saved AOI you want to work in.
4. Select the relevant label in Label List.
5. Choose Draw circle annotation.
6. Click the centre of an object.
7. Move to the object's edge and click again to finish the circle.
8. Repeat for every target object fully or partly inside that AOI.
9. Work through the remaining training, validation and test AOIs.

Use consistent labels and cover each object's visible extent as carefully as possible. A partially included object still needs an annotation; do not annotate only the easiest objects and leave other target objects unlabelled in the same AOI. Review crowded areas and objects touching AOI edges before training.

Circle annotation is the available drawing method in this interface. Do not plan a workflow around disabled tools. The Magic annotation and polygon annotation buttons are currently disabled.

Check that your work has saved

AOIs and annotations save automatically. New shapes can appear immediately while the save finishes. The training button may show Saving ... areas or Saving ... annotations. Wait for those pending saves to clear before training, leaving the page or closing the browser.

If a message says that a save could not be confirmed, reload the detector and inspect what is actually present before drawing it again. An interrupted response can make the outcome uncertain; immediately redrawing risks duplicates.

To remove an annotation, select it and use the delete-selection button or the Delete/Backspace key. To remove an AOI, select it using the appropriate selection tool and confirm deletion. Deleting an AOI also removes its annotations, so review the confirmation carefully.

Start and monitor training

1. Confirm the minimum 8/2/1 AOI counts.
2. Confirm there are saved annotations in the training, validation and test groups. Meeting that minimum is only a starting point: fully label each AOI.
3. Wait until pending area and annotation saves have cleared.
4. Click Train Detector.
5. If eligibility information appears, correct the stated requirement and try again.
6. Wait for Training queued, preparation, training, validation and model upload to finish.
7. On success, check that the detector is Active and review the reported accuracy.

Editing is locked during queued/running training so that the training inputs remain consistent. Progress can show epochs, the maximum epoch limit, the best epoch and an early-stopping count. Training can finish before the maximum because it may stop after enough epochs without improvement. The epoch-limit percentage is not an estimate of remaining time.

Accuracy is a model evaluation metric, not a promise that the same percentage of trees in every new survey will be correct. Review the model on imagery that represents your intended use.

After adding more annotations, Train Detector trains using all current AOIs and annotations, including existing examples, with the configured preset weights. It is not an instruction to train only on the newest circles. Successful training publishes the model and makes the detector active automatically; there is no separate activation switch to find.

If training fails, the dialog offers Cancel, View logs/details and Retry. Read the details before retrying unchanged data repeatedly.

11 Use Run on training TIFF

This action is a convenient way to assess a trained detector over its original image. It creates a separate inference project; it does not overwrite your AOIs or annotations.

1. Open the trained detector and wait for training to finish successfully.
2. Confirm it is active and the training imagery is available.
3. Click Run on training TIFF.
4. Review the Create inference project dialog. Its source is Use this detector's training TIFF and its model is initially the trained detector.
5. Edit Project name if desired.
6. Add optional Patch / boundary files to limit the assessment area.
7. Change Detection AI model if you want to compare another available model.
8. Click Process and respond to any boundary warning.
9. Follow the new project on Inference projects and review it using View when ready.

You do not need to choose the same TIFF from your computer again. The detector's original imagery is kept. Any newly selected boundary still needs to upload, so keep the browser tab open until submission is complete.

If the button is unavailable, read the nearby explanation. Common causes are training still running, no successfully published active model, or missing imagery. Credits are checked in the new project's settings and submission; insufficient credits can prevent processing after the settings dialog opens.

The new project starts its own paid-area record and follows inference retention. Free model changes within an older inference project's paid area do not make a newly created training-TIFF inference project free.

12 Retention deletion and keeping your own copies

Inference imagery and result files expire 14 days after the latest successful inference. The project displays its expiry using your browser's local time. A successful rerun starts a new retention interval; simply viewing, starring or downloading the project does not.

Download the orthophoto and output ZIP before expiry if you need them for reporting, GIS work or long-term records. After expiry, View, downloads and inference reruns are unavailable. To process the imagery again, upload it as a new project.

Detector projects, including their training data, remain until you delete them. Keep your own originals and any important exports: the application should not be your only archive.

To delete an inference project, open its three-dot menu, choose Delete and confirm. To delete a detector, use the detector card's menu and confirm Delete detector?. Both actions are intended to be permanent from the user's perspective.

Deletion removes the project from the interface and schedules removal of its private storage files. Storage cleanup may continue in the background. If deletion is refused, wait for the project's uploads or processing to finish. A detector can also be protected while queued or running inference is using its model.

Keep a simple local archive containing the survey source, final ortho, downloaded result ZIP, project name and date. For raw-image projects, remember that the original photographs are removed from application storage after the verified orthophoto is generated.

13 Troubleshoot common problems

When reporting a problem, provide the project or detector name, the action you clicked, the time, the exact visible error and a screenshot if useful. State whether the transfer, processing or preview failed. Do not share your password or private download links.

14 Two worked walkthroughs

Example A Count an existing block and compare models

You have a georeferenced North_Block.tif and a complete boundary shapefile representing 10 ha. Your account has 30 credits and you want to compare standard models.

1. Check that the TIFF is within 20 GiB and the boundary includes .shp, .shx, .dbf and .prj.
2. Create an inference project called North Block — model comparison.
3. Choose Orthophoto TIFF, select the image, add the boundary and choose your first standard model.
4. Click Process and keep the browser open until submission finishes.
5. Wait for Completed. If the measured usable boundary is 10 ha, the first successful run uses 10 credits and the balance becomes 20.
6. Download the first output ZIP and save it under a model-specific local name.
7. Open View, wait for the first preview if needed, and inspect boxes and centroids across representative locations.
8. Return to the list, change Detector selected to a second model, and run inference again using the same boundary.
9. The model-only rerun adds no area charge. Download its results and compare the two local outputs.
10. Save the required files before the displayed expiry date.

If you later expand the paid maximum to 12 ha using a standard model, the additional charge is 2 credits. If you instead expand from 10 to 12 ha using Nursery, the additional charge is 1,800 credits. Check the chosen model before expanding.

Example B Train from examples and test the whole image

You have one suitable training TIFF and want to build a detector from examples.

1. Create a detector called South Block detector and upload its training TIFF.
2. Wait for Imagery ready, then open Preparing training data.
3. Draw eight training, two validation and one test AOI using similar sizes and two clicks per rectangle.
4. Create a label, select the first saved AOI and use two-click circles to annotate every target fully or partly inside it.
5. Complete all AOIs, including the validation and test examples. Inspect crowded locations and boundary objects.
6. Wait for all pending saves to finish and click Train Detector.
7. Follow training, validation and publication until the detector is active. Review its accuracy as a model metric.
8. Click Run on training TIFF. Name the new inference project South Block — trained model review and add a boundary around the ground you need.
9. Review credits and click Process. The TIFF does not need a new local upload.
10. Open the completed inference View and compare detections against the image. Keep downloaded results and note where the detector misses objects or makes false detections.
11. If needed, return to the detector, add representative examples, wait for saves, and train again using the current dataset.

15 Checklists for everyday use

Before a new inference project

• I am signed into the intended account.
• I selected the correct TIFF or raw-image folders, and reviewed the combined image count and size.
• The files meet the size/count limits and filenames are unique.
• My boundary covers the intended ground and declares a usable CRS.
• I checked the selected model and its credit rate.
• I have a positive balance for a new project or added area.
• I can keep the tab open and computer awake until uploading finishes.

Before Train Detector

• The training image is ready.
• I have at least 8 training, 2 validation and 1 test AOI.
• AOIs have similar sizes and objects are about 10–20% of AOI width.
• Every target fully or partly inside each AOI is annotated.
• Labels are consistent and all three groups contain annotations.
• All pending area and annotation saves have finished.

Before closing a project or leaving the portal

• Uploads and any annotation saves have completed.
• I know whether the job is queued, processing, completed or failed.
• I downloaded the ortho and final output ZIP if I need local copies.
• I saved all shapefile components together.
• I checked the inference expiry date.
• I kept original photos outside the application before any ODM cleanup.

16 Glossary