3dsem

Classify point clouds with pretrained 3D semantic segmentation models, on your own machine, from one command.

pip install 3dsem

sem models
sem install [whichever model looks good to you]
sem output (to set a permanent output directory)

The classified .laz appears next to your input file, with per-point classification, confidence, and every original dimension carried over.

What you need

• An NVIDIA GPU with a current driver (Windows 527.41+, Linux 525.60.13+)
• Python 3.9 or newer

sem install shows the download size, the license, and a GPU check, then asks once before any bytes move. It is safe to interrupt and resumes where it stopped. After install, inference runs fully offline.

Models

dales-utonia is trained on DALES aerial LiDAR and predicts 7 classes: ground, vegetation, vehicle, powerline, fence, pole, building.

dales-hag-utonia predicts the same 7 classes from the same data, with height above ground added as an input. It scores higher on held-out DALES scenes (0.86 mIoU against 0.98 overall accuracy) and is the one to reach for first.

ieee-utonia is for low density airborne LiDAR and predicts 5 classes: ground, vegetation, building, water, bridge deck. Alongside geometry it reads return number, and the run it was trained from recorded that channel by column position, so its name (column_4) says nothing about what it holds. sem binds it for you off the return number your file names, which a LAS or LAZ always does. A .ply or .pcd that names it differently needs the binding by hand:

sem infer ieee-utonia scan.ply --input-field column_4:return_number

h3d-utonia is for coloured photogrammetric clouds and predicts 8 classes: ground, vehicle, urban furniture, roof, wall, low veg, tree, chimneys. It is the one model that consumes colour, so it needs --rgb-max naming the full-scale value your file uses:

sem infer h3d-utonia tile.laz --rgb-max 255

Every model is licensed CC-BY-NC 4.0 (non-commercial)

sem models lists what is available and installed. Bare sem opens an interactive picker.

How do I make it more accurate?

Use a preset. Each one turns on more of the same three ideas: predict the scene from several augmented views and vote (test-time augmentation), tile the scene a second time at a half offset so no point sits on a tile edge (overlapped voting), and clean up the labels afterwards (smoothing, island removal, geometry rules).

• default: one pass with light smoothing
• --med: 4 voting views, island removal, single tiling pass. Roughly 4x.
• --high: 6 views including flips and rotations, stronger smoothing, and the overlapped second pass. Roughly 24x, because the views and the second pass multiply.
• --ultra: 9 views, the strongest smoothing, overlapped. Roughly 36x.

How to evaluate accuracy on labeled dataset?

sem eval tile_pred.laz tile_truth.laz

No inference runs. All it does is reads the Classification column of both files and prints, per class, IoU, precision, recall and F1 with point counts, then overall accuracy, mIoU, macro F1, Cohen's kappa, and a confusion matrix. Files written by sem infer carry their own class legend, whatever codes that run actually used, so rows read 2 ground rather than bare numbers. A file without a legend shows plain codes, and --names 1=ground,4=water names any code by hand. The files must be the same points in the same order, which is what sem infer's own export gives you. --pred-field and --truth-field name a different column on either side.

When the truth file uses different codes than the model, --map translates them first, and --ignore keeps codes out of the scoring:

sem eval tile_pred.laz tile_truth.laz --map 3=5,4=5 --ignore 0,1

folds the truth's low and medium vegetation (depicted as 3 and 4 in the classification field) into the model's single vegetation class and skips points that were never classified.

To see whether the cleanup passes help or hurt on your data, run the inference with --keep-raw: it also exports a copy from before post-processing into a raw/ folder inside the job, under the same names. Score each against the same truth and compare:

sem eval tile_dales-hag-utonia_predictions/ truth/
sem eval tile_dales-hag-utonia_predictions/raw/ truth/

A whole batch works the same way with two folders: files pair by name (tile_pred.laz matches tile.laz), each tile gets a summary line, and the totals pool every point before computing IoU, so a big tile counts for more than a small one. A predicted file with no truth partner stops the run and is named. A folder is evaluated against a folder, a file against a file.

How do I make it faster?

--low is a single pass with no cleanup. If a preset is mostly what you want, --no-overlap drops its second tiling pass, which is about half its extra inference time.

Big buildings come out patchy or cut through. Why?

The scene is processed in square tiles, by default 50 m on a side. An object bigger than one tile is predicted in pieces, and the pieces can disagree. Two fixes that combine well:

• --chunk-xy 100 makes the tiles bigger, so a large building fits in one. Costs GPU memory.
• --overlap (on automatically with --high and up) predicts a second pass at a half offset and votes, which removes most seam artifacts. It costs about 4x on its own.

It ran out of GPU memory

Lower --chunk-xy, try 35 and then 25. Smaller tiles need less VRAM, and the extra seams they create are what --overlap is for.

Poles or powerlines are disappearing

Presets from --med up turn on island removal, which absorbs clusters smaller than 10 points into their surroundings. Thin objects are exactly small clusters. Keep the filter but make it gentler with --sieve-min-pts 5, or turn it off with --no-sieve.

Can I hide the model's low-confidence guesses?

--unclass 0.6 exports every point below 60% confidence as unclassified instead of its best guess. Bare --unclass uses 0.5. To see where the model is unsure, --extra-dims adds per-point diagnostic fields: entropy (0..1 uncertainty), margin (gap between the top two classes), probs (one probability field per class, grows the file), and ood (the raw scores behind the gates below). Combine them freely: --extra-dims entropy,margin.

It confidently labels things it has never seen

A crane or a boat has no class to land in, so the model puts it in the nearest one it knows, often with high confidence. Confidence alone will not catch that, because the model is confident and wrong. Two other scores catch it instead.

--unclass-gmm compares each point against how the training classes actually looked to the model and unclassifies anything too far from all of them. Bare, it uses the threshold the model was shipped with; give it a number to override. --unclass-maxlogit 4.0 catches the opposite case, points where no class drew much evidence. They find different mistakes, so using both is normal.

To choose your own numbers, run once with --extra-dims ood, which writes the raw scores into the output file so you can see where your data sits, then set the thresholds and re-export. Re-exporting does not re-run the model.

Part of my file is already classified correctly

--input-field preserve:Classification:2,9 keeps the codes your file already carries for those points and writes the model's prediction everywhere else. They are read from the source file at export, so inference, the unclassified gates and the cleanup passes never touch them.

Kept codes are written as your file spells them. When a kept code means something different in the model's output (your water 9 has no model class), add --preserve-map 9=2 to write it as ground instead. The export warns whenever a kept code is not one the model uses.

The export reports how often the model disagreed with what you preserved, which is a quick check on both.

You can also do it afterwards, on a file that is already classified:

sem preserve tile_pred.laz tile.las --keep Classification:2,9

That writes tile_pred_preserved.laz beside it. Same overlay, decided after you have seen the result rather than before, with the same --map 9=2 option for codes that need translating. The two files have to be the same points in the same order, which is checked on counts and coordinates. The predicted file needs to be on ASPRS codes for this: the codes you keep come from your own file's scheme, so overlaying them onto raw model indices would put two vocabularies in one column. Export it with --asprs, which preserving during inference turns on for you.

Can I combine models?

sem infer dales-utonia+dales-hag-utonia tile.las runs every model in the chain and merges their predictions with a vote: each model's per-class probabilities are averaged, and the strongest combined evidence wins. Exact ties go to the model you listed first, so lead with your strongest.

The models do not have to predict the same classes. Classes spelled the same way are one class that every model votes on; a class only one model has is that model's word, and the models that never saw it do not get to vote it down. So

sem infer dales-hag-utonia+ieee-utonia tile.laz

gives you nine classes: DALES' seven, with water and bridge deck added from ieee-utonia. The run prints the combined legend and who contributed to each class before any model starts, so a mix that is not what you meant costs nothing to abandon.

The result carries an agreement field (what fraction of the models entitled to name that class agreed) and an ens_member field (which model drove each label). Each member's own predictions sit in members/ inside the job, so one run leaves one folder and you can still compare them individually.

The models call the same thing by different names

Classes only merge when they are spelled alike, so dales-hag-utonia's vegetation and h3d-utonia's tree stay separate and compete. --merge folds them into one class:

sem infer dales-hag-utonia+h3d-utonia tile.laz --rgb-max 255 \
    --merge vegetation="tree,low veg" \
    --merge building="roof,wall,chimneys"

Repeat it per class. Everything you do not mention still merges by name, so you only write down the parts sem cannot know.

--merge names a class of the result and the member classes that feed it. It only goes that way round: several member classes can become one, but one cannot be split into several, because a model that never learned the difference has no opinion about which side a point falls on.

Classes you leave out of a merge are still classes. Nothing is dropped.

Can I run my own trained model?

If a training run left you a folder holding final_model.pth and run.json, register it:

sem local-model C:\runs\my_run --name myrun

It appears in sem models and runs like any other: sem infer myrun tile.las. Nothing is copied; the folder is used in place, and sem local-model --forget myrun unregisters it without touching the files.

Inference borrows the installed env of a catalog model with the same backbone, so install one that matches first. Everything else the run needs (grid, features, classes) is read from the run.json beside the weights.

My file has no coordinate system

It still runs. sem carries the CRS through to the classified output when your file has one and writes none when it does not. Coordinates are used as stored; nothing is reprojected.

The unit is taken on trust. Tile size, voxel grid and neighbourhoods are lengths in metres, so a file in metres is correct and a file in feet is off by that factor. Most aerial LiDAR is already metre-projected. If yours is not, convert it first, or state the lengths in your file's own unit with --grid and --chunk-xy.

--epsg 26917 sets the CRS written to the output (use your own zone's code). It does not transform anything.

Coordinates in degrees are worth watching. A 50 m tile covers a fraction of a degree several times over, so the whole scene lands in one tile. sem says so and continues; reproject to a metric CRS first if the result looks wrong.

My file's coordinate system is wrong

A LAS header can be absent, or present and wrong. Exports that record the datum and drop the projection are common, and the file then claims degrees while holding metres.

This does not affect the classification, since the CRS only labels the output. --epsg overrides whatever the header says.

My file is not a LAS. How does sem know what its columns mean?

It does not. LAS and LAZ name their dimensions in the format spec, so intensity, return number and classification are read straight off a LAS with no help from you. .ply and .pcd leave the meaning of a property up to whoever wrote the file, so sem asks you:

sem infer dales-utonia scan.ply --input-field intensity:scalar_Intensity

--input-field KEY:COLUMN[:VALUE] says where one thing lives in your file, repeated per entry. Columns are names or 0-based numbers.

• xyz:A,B,C and rgb:R,G,B take three columns
• intensity:C, return-number:C and hag:C take one
• ground:C:V,... and preserve:C:V,... add the value(s) in that column that mean it, since the column alone does not say which value counts

A model that wants a channel you did not name stops and lists the fields your file actually has.

Nothing is inferred from a column's position or from a name that looks familiar, because a file with id,x,y,z in that order and a file with x,y,z,id are indistinguishable to anything except you.

My colours come out black, or sem asks for --rgb-max

Point clouds store colour as 8, 10, 12 or 16 bit, and no format records which. --rgb-max states the full-scale value: 255, 1023, 4095, 65535, or 1 for float 0 to 1 colour. It is only needed when the model actually consumes colour; a model that runs on intensity ignores the colour in your file and never asks. Of the shipped models that is h3d-utonia alone.

On a las or laz you do not have to supply it. sem reads the colour before the run and takes the smallest of those values that covers the brightest point it saw, printing what it picked:

[portal] color: values up to 4032 -> --rgb-max 4095

Read that line, because it comes from the first million points rather than the whole file. A tile whose bright points all sit late in it reads low, and a 12-bit cloud scored as 10-bit comes out dark. Passing --rgb-max yourself overrides the probe. Every other format has no colour sem can probe this way, so it stops and asks.

Height above ground

Models trained with a HAG channel compute one at staging, using a ground raster whose cell size comes from the linear unit your file's CRS declares: 2 metres, or whatever length equals 2 metres in your file's own unit. A cloud in US survey feet gets a 6.56 foot cell, which is the same ground resolution.

A cloud that declares no projected CRS has no unit to read, so the cell stays 2, which is right if the coordinates are metres and wrong by that factor if they are not. --epsg supplies a CRS to read the unit from, and --hag-cell states the cell outright and skips the question.

--hag-cell also overrides the derived cell when you want a different resolution. The height error it costs is roughly the cell size times the local slope, so a smaller cell buys accuracy on steep ground and costs memory.

--ground-method picks where the ground comes from (smrf, csf, zmin, or labels when your file already marks ground). --csf-rigidness (1 steep, 2 moderate, 3 flat/urban), --smrf-window and --smrf-cut tune the ground filters. If your file marks ground with a class code, say so in one entry: --input-field ground:Classification:2. The column alone is not enough, because it does not say which value in it means ground. More than one code can count as ground: ground:Classification:2,9 also folds water under the ground surface, which keeps HAG flat across lakes and rivers.

The cleanup rules are wrong for my data

The geometry rules only run when you ask for them and tell them what your classes mean: --rules --roles veg=vegetation,building=building. sem no longer decides that a class is vegetation because its name contains "tree". Their thresholds are lengths in your scene's vertical unit and unitless ratios, all settable through --rule-set: ground_hag (0.1), lowveg_hag (0.35), high_hag (2.0), planar_min (0.55), scatter_min (0.4). For example --rule-set high_hag=3,planar_min=0.6.

Controlling the exported class codes

By default each class exports as its own model index: 0, 1, 2 and so on, in the order the model lists them. On the shipped DALES models that is ground 0, vegetation 1, vehicle 2, powerline 3, fence 4, pole 5, building 6, and gated points 7.

Those indices are not ASPRS codes. In the spec 2 is Ground, so a viewer reading your file per the spec sees the vehicles as ground. Exported las and laz carry a legend recording what each code meant, and sem eval reads it, but nothing outside sem does. Any file leaving sem wants the next flag.

--asprs exports spec codes instead: ground 2, vegetation 5, building 6, powerline 14, pole 15, and user-definable codes from 64 for classes the spec has no name for, which on these models is vehicle 64 and fence 65. A class the table does not name takes the next free code at or above 64.

--asprs-map ground=2,building=6 overrides any of them by name and implies --asprs. Two classes landing on one code stops the run rather than merging them silently. Preserving codes from your own file implies --asprs too, since those codes already speak it.

--unclassified-code sets the code gated points receive. It defaults to the class count, or to 1 under --asprs, which is ASPRS "Unclassified" and no class maps onto. Export stops rather than letting that code collide with a real class.

I don't have a GPU

sem classifies on your own machine and needs an NVIDIA GPU. Conversion and export are CPU work; only the model needs one.

If you do not have one, --modal rents one for the length of a run:

pip install "3dsem[modal]"
modal token new

sem infer dales-hag-utonia tile.laz --modal

modal token new opens a browser and links your Modal account. It is asked once. After that, --modal on any sem infer uploads your file, runs the same command on a cloud GPU, and writes the results where a local run would have put them: next to your input, or under sem output if you set one.

Every flag works. Presets, --asprs, --unclass, --input-field, model chains, a whole folder of tiles.

The first run installs the model on Modal, which takes a few minutes. Later runs reuse that install and start in about a minute. Modal bills for run time only, so an idle account between jobs costs nothing.

SEM_MODAL_GPU picks the GPU (default A10G) and SEM_MODAL_TIMEOUT_HOURS caps one run (default 4).

Your own weights go up too. sem infer myrun tile.laz --modal uploads the folder you registered with sem local-model and runs it against a catalog model's environment, so a machine with no GPU can run a model it could not install.

One limit: --modal needs Python 3.10 or newer, where the rest of sem runs on 3.9. Your Python and the rented machine's have to match.

Is my data safe on Modal?

Your file is uploaded to a private Modal volume in your own account, and the uploaded copy is deleted once the results come back. Results stay on the volume until you remove them. Nothing is shared with anyone, and nothing is sent anywhere else.

If that is not acceptable for your data, classify locally. The models are the same either way.

Where do Modal jobs go?

sem output-modal shows what is in use:

sem output-modal            show the volumes and the GPU
sem output-modal site-a     send jobs to the site-a volume
sem output-modal off        back to sem-outputs

This is separate from sem output. One names a Modal volume, the other a folder on your disk, and both apply to a --modal run: the job is written to the volume and a copy comes back to the folder.

Two volumes are used, both created on first use in your own account. sem-outputs (or whatever you named) holds the jobs. sem-envs holds the installed models, which is why the first run is the slow one. modal volume delete sem-envs clears them and the next run reinstalls; modal volume ls sem-outputs lists what jobs are still up there.

A Modal run failed partway. What now?

Re-run it. Nothing is left in a state that needs cleaning up first.

When a run fails, whatever it managed to write stays on the volume under the timestamp printed in the error, and your uploaded input stays with it, since sem only clears the upload after results come back safely. Both are visible with modal volume ls. Re-running uploads afresh under a new timestamp and does not touch the old one, so modal volume rm sem-outputs <timestamp> -r is how a failed attempt goes away once you no longer want it.

Can I rerun with different settings without reconverting?

Yes, if you pass --keep-npz on the first run.

The output folder is a self-contained job, named after your file and the model (tile_dales-utonia_predictions/). The staged .npz inside it hold the converted channels, and they are what a re-run reuses. By default they are deleted once the export is written, because most runs only want the classified file. --keep-npz retains them.

With them kept: quality options (presets, TTA, cleanup, export) never reconvert. Conversion options (--epsg, --ground-method, --hag, ...) and changes to the input file itself reconvert automatically; identical settings reuse the staged files. Every model keeps its own job folder, so switching models never mixes results.

Can I just name the output file?

Yes. An output ending in .laz, .las, .txt, .csv or .ply is the file itself, and its extension picks the format, so --format becomes redundant:

sem infer dales-utonia tile.las out/pc.laz

The job folder is created beside it (out/pc_dales-utonia_predictions/), so everything the run makes stays where you pointed. This is for a single input file; a folder of tiles produces one result per scene, so give it a folder.

If you have set a default with sem output, a bare filename lands there: sem infer dales-utonia tile.las pc.laz writes <your dir>/pc.laz. Put a folder in the path and it goes exactly there instead.

Where does everything live?

Downloads go to ~/.trainer (set TRAINER_HOME to move them). Your data and results never go there: each job is a folder next to your input, or under a default you set with sem output <dir> and undo with sem output off. sem clean dales-utonia removes one model; sem clean --all removes everything sem ever downloaded.

Can I look at the intermediate files?

Run with --keep-npz, or the intermediates are gone once the export is written. Then sem tolaz job/tile_input.npz writes tile_input.laz beside it. The xyz becomes the cloud, rgb the color, and every other per-point channel a named field you can shade by in CloudCompare. That works on the staged input as well as on predictions, so it is how you see the features a model actually received, not just what it predicted.

Can I build my own preset?

--pick gives you the six knobs the presets are made of, on arrow keys:

sem infer dales-hag-utonia tile.laz --pick

your settings
  tta            -
  overlap        default
  no-postproc    off
  sieve          default
  knn-k          -
  knn-radius     -
  start the run

Set the ones you want and leave the rest. A preset is a name for one combination of these; here you choose your own.

The command builds itself in the footer as you go, and is printed before the run, so this is also a quick way to learn the flags:

sem infer dales-hag-utonia tile.laz --tta full --sieve --knn-k 20

You can start from a preset. sem infer m tile.laz --med --pick opens with the knobs showing what --med means, and anything you change is added to it:

sem infer m tile.laz --med --knn-k 20

Anything else you pass alongside --pick is carried through untouched, so sem infer m tile.laz --asprs --pick keeps --asprs.

Every option

sem infer --help documents all of it, and sem infer --help all shows every flag in one table.

Licensing

The sem tool is MIT licensed. Each model ships a NOTICE.md stating its architecture credits and license terms; some models carry a non-commercial restriction inherited from their pretrained components, shown before you install.