Frame2FormCAPTURE · UPLOAD · RECONSTRUCT

Getting started

Frame2Form builds textured 3D models from your own footage, in your browser, with nothing to install.

Your account

Accounts are by invitation. The invitation comes from no-reply@verificationemail.com, with the subject "Your Frame2Form account" and a temporary password valid for 7 days; check your junk folder if it has not arrived. Press Sign in on Frame2Form (frame2form.ghz599.net), use your email address and that password, then choose your own: at least 12 characters, with a capital letter, a small letter and a number.

What you need

• A computer on broadband or Wi-Fi for the upload. A drone video can be 3 GB; 57 phone photos came to 173 MB. Keep the page open until the upload finishes. Following a job and viewing models work on a phone.
• For a video, the file exactly as it came off the drone. An edited, exported or shared copy has lost the hidden flight record, and is refused.
• For a video, if you can, the ground height where you took off, in metres above sea level. The service suggests one; your own figure is better.

How a job works

Every job, from a video or from photos, goes through the same five steps. Only the first and third need you at the computer.

Times grow with the size of the video and the number of photos.

From a drone video

Only the DJI Neo 2 so far: its hidden flight record has been decoded and checked against test recordings. Other drones' videos are not supported yet.

Flying for a good model

• Round a building, statue or monument: orbit it at a steady distance with the camera kept on the subject. A second orbit at a different height, looking further down, fills in roofs and ledges.
• Along a facade or wall: fly slowly along it at a steady distance, facing it, then again a little higher or angled. Carry on round corners rather than stopping at them.
• Turn slowly. Fast turns blur the frames; the service leaves them out, and the model has gaps.
• Wait for GPS before take-off. Without positions the frames cannot be geotagged and the job stops.
• Keep the subject filling much of the view, and see it from every side you want in the model: what the camera never saw cannot be built.
• Prefer even light: an overcast day is better than low sun with hard shadows.

Uploading and confirming

1. On New job, choose Drone video, press Choose video and pick the .MP4, add a name if you like (for example "St Mary's shrine, west side"), and press Upload. When the upload finishes, the job's own page opens.
2. After about 3 minutes it shows Waiting for you, with a banner on Jobs and a count on its tab. Press Confirm to open the map: the blue line is the flight and the dots the chosen frames; green marks the take-off point; red is the subject; the orange outline is the suggested boundary; green shading is where the camera looked.
3. Enter the take-off ground height, or press one of the suggestions (most accurate first: in England, the Environment Agency's 1 m LiDAR). It sets the height above sea level of every frame. It is needed for a model; for frames only you can leave it blank, and then no height is written.
4. Choose Frames only or Frames and model. For a model, the model is built only inside the orange outline: drag a corner to move it, click a corner to delete it, or press Draw my own and click out a new outline (click the first corner to finish); Reset boundary goes back to the suggestion. The shaded area is always kept, so a tight outline cannot cut the subject off.
5. Press Start. You can close the page.

From photos

Taking the photos

The software finds the same points in many photos taken from different places, and works out the shape from how they shift. So every part of the subject needs to be in several photos, each taken from a slightly different position. Move between photos; never turn on the spot.

• An object, statue or small building: walk round it, a photo every 10 to 15 degrees (about 30 a circle), the subject filling most of the frame. Go round again higher or lower, then add closer photos of detail, moving in gradually.
• A facade or wall: walk along it, a photo every third of a frame width; one pass straight on, one angled left, one angled right. Go round corners in small steps, and step back for a few photos of the whole front.
• A room, from inside: move round the edge facing the opposite wall, stepping sideways between photos. Photos from one spot make a panorama, which has no depth, and the model fails. Plain walls, windows and mirrors give the software little to match.
• For any subject: keep the zoom fixed (on a phone, the main 1x camera); hold still for each photo; prefer even light; avoid moving people, glass, shiny or very plain surfaces.
• How many: 50 to 300 suits most subjects; at least 20 are needed.

Uploading and confirming

1. On New job, choose Photos, then Choose photos (photos, or one ZIP of them) or Choose folder. Add a name if you like and press Upload photos; when the upload finishes, the job's own page opens. Files that are not JPEG or HEIC (raw files, videos, iPhone edit files) are left out, and the page says how many. Raw files must be exported to JPEG first.
2. After a few minutes the job shows Waiting for you. Press Confirm: the summary lists the photos used, any left out and why, the cameras found, whether the photos' GPS positions will be used, and any warnings such as slow shutter speeds.
3. Choose the resolution. Full uses the photos as taken; Reduced (4,000 pixels on the long side) is faster and fits more photos. Each shows an estimate of time and cost. If you used more than one camera or zoom setting, you can use only the largest group. When positions are used, a map shows them and you can draw a smaller area.
4. Press Build model. You can close the page.

Following a job

Jobs lists your jobs, newest first, one line each, and updates itself while anything is running. Each line shows the job's status, a short note and the one thing to do next; the buttons above the list show only jobs in progress, ready, or needing you, and Find a job searches by name. Press a job to open its own page.

A job's own page

It has everything about one job: its status and what the service said, with Confirm, View model or Retry on the larger server when they apply; Summary (for a video, Summary and map), which reopens the map or photo summary at any time; the Downloads, each with a line on what it is for; and down the side, Sharing, Details (take-off height, textures, and when the upload and the results will be deleted) and Delete, which removes the upload, the results and the job (not while it is running).

Viewing a model

Press View model. Drag to turn, scroll or pinch to zoom, right-drag or drag with two fingers to move; Reset view puts everything back as it was when the viewer opened (the starting view, Colours, the light over your shoulder, Triangles off; Full textures stay loaded). ? in the viewer shows a short version of this list beside the model.

The viewer always shows every face of the mesh. Full textures, or failing that the Model ZIP, has the textures at their original size. Models from drone video have no Full textures: video frames hold no more detail than the viewer's copy already shows.

Trimming a model

Trim cuts a finished model down to the part you want (a statue without its plinth and surroundings, say) in about a minute, without building it again. Any finished model can be trimmed.

1. In the viewer, press Trim. The view turns to look straight down.
2. Click round what you want to keep, corner by corner; click the first corner to finish.
3. Set Bottom and Top with the sliders. What will be cut away is dimmed at once; turn the model to check it from the sides. Redraw outline starts again.
4. Press Make trimmed copy. The job shows a Trim line; when it reads Ready, the trimmed files appear under their own heading in the downloads, and View model opens the trimmed copy, with Show whole model to switch back.

The whole model is never changed; trimming again replaces the previous trim. Only the model is cut: the point clouds and the full OpenDroneMap output stay whole. For a drone model the outline is also saved on the map (trim.geojson in the Trimmed model ZIP).

Names and sharing

Names

Give a job a name when you upload, or later with Rename beside the name on the job's page. The name is shown in place of the file name, which stays underneath in grey. Leave it empty to go back to the file name.

Sharing a model

On a finished model's page, under Sharing, press Share (or Change sharing) and choose:

• Not shared: only you can see it.
• Everyone: every Frame2Form user can view and download it.
• Specific people: type the email address of each person's Frame2Form account and press Add (up to 20). An address without an account is refused, and nothing changes until every address is right.

They find it on their Shared page, marked "Shared with everyone" or "Shared with you", with your email address as who shared it, and can view it, see its summary and download it. Only you can change the sharing or delete it, and deleting removes it for them too. The original video and the prepared photos are never shared.

Downloads

The downloads are on each job's page. For the model, the Model ZIP is what most people want. Results are kept for 30 days; the date is under Details on the job's page. Sizes are from the 3 GB test video and the 57-photo test.

Limits and how long things are kept

• Uploads are deleted after 7 days and results after 30. Download anything you want to keep.
• 10 models a month per user, from videos and photos together.
• Up to 1,000 photos or 10 GB in one upload.
• Positions come from the drone's GPS, not a survey: about 3 m across and in height. The shape and size are far more reliable than the position. Heights are measured from the take-off figure, so an error there moves the whole model up or down.
• Most photo models have no scale or position: the right shape, in its own units, not facing north and not on a map. Scale it yourself from a known length.
• Models are checked, not perfected. The build report counts faces that could not be coloured and small holes, rather than hiding them.

When something goes wrong

Help

For questions or help, email mark@willowhey.net, with the job number if it is about a job.

Frame2Form is built with open-source software and open data; see Software credits.