GRASS logo

NAME

i.sam3.opencl - Segments a 3-band imagery group with the Segment Anything Model 3 (SAM 3) on an OpenCL device and writes the segments as vector areas.

KEYWORDS

imagery, segmentation, object recognition, deep learning, GPU, OpenCL, vectorization

SYNOPSIS

i.sam3.opencl
i.sam3.opencl --help
i.sam3.opencl [-n] group=name [subgroup=name] [checkpoint=name] [weights_dir=name] output=name [raster=name] [percentile=lower,upper] [prompt=text[,text,...]] [threshold=float] [points_per_side=integer] [iou_threshold=float] [stability_threshold=float] [stability_offset=float] [nms_threshold=float] [crop_layers=integer] [crop_points_downscale=integer] [minsize=integer] [maxsize=integer] [overlap=string] [simplify=float] [tile_size=integer] [tile_overlap=integer] [batch=integer] [device=string] [platform=string] [profile=string] [precision=string] [--overwrite] [--help] [--verbose] [--quiet] [--ui]

Flags:

-n
Do not download a missing checkpoint into the weights folder
--overwrite
Allow output files to overwrite existing files
--help
Print usage summary
--verbose
Verbose module output
--quiet
Quiet module output
--ui
Force launching GUI dialog

Parameters:

group=name [required]
Name of input imagery group with exactly 3 raster maps (R, G, B)
subgroup=name
Name of subgroup with exactly 3 raster maps to use instead of the whole group
checkpoint=name
SAM 3 checkpoint file
Original PyTorch checkpoint sam3.pt
weights_dir=name
Folder of the SAM 3 checkpoint
Folder holding sam3.pt, where it is downloaded if missing; default: $SAM_MODELS_DIR/sam3, else ~/models/sam/sam3
output=name [required]
Name for output vector map of segments
raster=name
Name for optional output raster map of segment categories
percentile=lower,upper
Percentiles of each band stretched to the 0-255 range fed to SAM 3
Options: 0-100
Default: 2,98
prompt=text[,text,...]
Text prompts: objects to search for
Short English noun phrases (e.g. building,swimming pool); every instance found is a segment. Without prompts the region is segmented automatically from a grid of point prompts
threshold=float
Minimum detection score of an instance found by a text prompt
Options: 0-1
Default: 0.5
points_per_side=integer
Number of prompt points along each side of a crop
Options: 1-128
Default: 32
iou_threshold=float
Minimum mask quality predicted by the model
Options: 0-1
Default: 0.8
stability_threshold=float
Minimum mask stability under a shift of the logit cutoff
Options: 0-1
Default: 0.95
stability_offset=float
Logit cutoff shift used to compute the stability score
Default: 1.0
nms_threshold=float
Box IoU above which duplicate masks are suppressed
Options: 0-1
Default: 0.7
crop_layers=integer
Number of zoomed-in crop layers (layer n has 4^n crops)
Options: 0-3
Default: 0
crop_points_downscale=integer
Divide points_per_side by this factor in each crop layer
Options: 1-8
Default: 1
minsize=integer
Minimum segment size in cells
Smaller holes and islands are removed from masks and smaller pieces merged into their neighbours
Options: 0-
Default: 100
maxsize=integer
Maximum mask size in cells
Larger masks, typically a background mask spanning the whole scene, are discarded
Options: 1-
overlap=string
Mask kept where masks overlap
Options: small, large, score
Default: small
small: smallest mask (finest partition)
large: largest mask (coarsest partition)
score: mask with the highest predicted IoU
simplify=float
Boundary simplification tolerance in map units
Douglas-Peucker tolerance; default is one cell, 0 keeps the raster staircase
Options: 0-
tile_size=integer
Side of the tiles, in cells, a larger region is split into
Options: 64-
Default: 1024
tile_overlap=integer
Overlap between tiles in cells
Options: 0-
Default: 256
batch=integer
Number of point prompts decoded at once on the device
Options: 1-256
Default: 32
device=string
OpenCL device type
Options: auto, gpu, cpu
Default: auto
platform=string
Use only OpenCL platforms whose name contains this text
profile=string
Device profile choosing buffers and kernels
Options: auto, cpu, igpu, dgpu, egpu
Default: auto
auto: Detect from the device
cpu: Host-memory device (PoCL): mapped buffers, CPU kernels
igpu: GPU sharing host memory: mapped transfers
dgpu: GPU with its own memory on internal PCIe
egpu: GPU behind a slow link (Thunderbolt, remote): pinned staging
precision=string
Storage type of the weight matrices
Options: fp32, fp16
Default: fp32
fp32: Single precision
fp16: Half precision storage, single precision arithmetic

Table of contents

DESCRIPTION

i.sam3.opencl segments a 3-band imagery group with the Segment Anything Model 3 (SAM 3) and writes the segments as a clean vector area map. The network runs on an OpenCL device (GPU or CPU), with weights read directly from the original PyTorch checkpoint file: neither Python nor PyTorch is needed at run time.

The module works in one of two ways:

In both cases the overlapping masks are then turned into a non-overlapping, topologically clean vector map with one attribute row per segment. SAM 3's video tracker is not used.

The group (or its subgroup) must contain exactly three raster maps, which are used as the red, green and blue channels in the order they are listed in the group (see i.group). Each band is stretched linearly between its lower and upper percentile to the 0-255 range SAM 3 was trained on. Cells that are NULL in any band are not segmented. Processing is done in the current computational region.

Checkpoint

The checkpoint is the original sam3.pt (3.2 GiB) published by Meta on Hugging Face (https://huggingface.co/facebook/sam3). The model is gated: an account must request access there, and downloading needs that account's access token.

Without a checkpoint file, the module looks for sam3.pt in the weights folder: weights_dir, by default the central folder $SAM_MODELS_DIR/sam3, or ~/models/sam/sam3 when the environment variable SAM_MODELS_DIR is not set. If it is not there, it is downloaded into that folder (created if needed) on first use and reused afterwards. The download uses the token in the environment variable HF_TOKEN, else the one saved by hf auth login (~/.cache/huggingface/token); it is checked against the published size, resumes where an interrupted one stopped, and waits for another run downloading the same file. The -n flag forbids downloading: a missing checkpoint is then an error. Checkpoints of SAM and SAM 2 are refused: they belong to i.sam.opencl and i.sam2.opencl.

Text prompts

Each prompt is a short English noun phrase in plain ASCII text (several are given comma-separated). SAM 3 encodes it with its text encoder, fuses it with the image features, and its detector proposes 200 candidate objects per crop. A candidate is kept when its score, the model's probability that it matches the prompt multiplied by the probability that the concept is present in the crop at all, exceeds threshold (0.5 by default, as in SAM 3's own Sam3Processor); its mask is the detector's segmentation. Crops are processed as in automatic segmentation (tile_size, tile_overlap, crop_layers), masks cut by an inner crop edge are discarded, and duplicates found in overlapping crops are removed (box IoU above 0.7), separately for each prompt. The options of the point-grid generator (points_per_side, iou_threshold, stability_threshold, nms_threshold) are not used.

Prompts are matched as SAM 3 was trained: on short, concrete noun phrases. A segment belongs to the prompt that found it (the prompt column); where masks of different prompts overlap, the overlap rule decides, with score using the detection score.

Mask generation

The region is processed in crops. A region no larger than tile_size cells is one crop, a larger region is split into tiles of tile_size cells overlapping by tile_overlap cells. With crop_layers greater than 0, each tile is additionally segmented in zoomed-in sub-crops (4 in layer 1, 16 in layer 2, ...), prompted with points_per_side divided by crop_points_downscale per layer.

Each crop is resized to 1008 x 1008 pixels, regardless of its aspect ratio, as SAM 3 does (the 8-bit antialiased bilinear resize of torchvision, reproduced to the bit), and passed through the image encoder. The mask decoder is then prompted with a grid of points_per_side x points_per_side points, batch prompts at a time, returning three candidate masks per point. A mask is kept if its predicted IoU exceeds iou_threshold and its stability score (the IoU of the mask thresholded at plus and minus stability_offset) is at least stability_threshold. Duplicates whose bounding boxes overlap by more than nms_threshold (IoU) are suppressed, keeping the best one. Masks touching an inner tile or crop edge are discarded, since the neighbouring tile sees them whole. Holes and islands smaller than minsize cells are finally removed from each mask.

As in SAM 3's point predictor, holes of at most 256 pixels in each 288 x 288 low-resolution mask are filled before the mask is resized to the crop. The other steps and their defaults are those of SAM 2's SAM2AutomaticMaskGenerator (SAM 3 has no generator of its own), plus the full-resolution removal of holes and islands smaller than minsize cells. With minsize=0, the module returns the same masks as those steps run on the PyTorch SAM 3 predictor.

Vectorization and cleaning

SAM 3 masks overlap: a field and the parcels inside it can both be masks. The overlap rule chooses which mask a cell belongs to: the smallest covering mask (small, finest partition, the default), the largest (large) or the one with the highest predicted IoU (score). Masks larger than maxsize cells are discarded first; this is the way to drop the single mask SAM 3 often returns for the whole background of a scene.

The resulting label raster is cleaned: every connected piece smaller than minsize cells, including small unsegmented gaps, is merged into the neighbouring piece it shares the longest border with. Boundaries are traced along cell edges into shared topological boundaries, simplified with the Douglas-Peucker algorithm using the simplify tolerance (one cell by default; 0 keeps the raster staircase), and the topology is cleaned (boundaries broken at intersections, duplicates removed, pseudo-nodes merged, slivers smaller than one cell removed). Each segment gets one centroid, placed at the cell farthest from its border.

Areas without a centroid are gaps that no mask covers. The optional raster output holds the segment categories of the cleaned label raster before simplification.

Attribute table

Column Description
cat Segment category
mask Number of the SAM 3 mask the segment comes from
predicted_iou Mask quality predicted by SAM 3
stability Stability score of the mask
cells Segment size in cells
area Area of the simplified polygon in square map units

With text prompts, the table holds the columns cat, mask, prompt (the text prompt that found the segment), score (its detection score), cells and area.

A mask split into several pieces by overlapping masks yields several segments with the same mask number.

NOTES

Differences from SAM and SAM 2

SAM 3's image encoder is a 32-block vision transformer (Perception Encoder, 1024 channels, 2D rotary position encoding, windowed attention with four global blocks) trained jointly for concept detection and tracking. Its point-prompt head is SAM 2's mask decoder on a 72 x 72 embedding of a 1008-pixel input, fed with 288 and 144 pixel high-resolution features from a SimpleFPN neck. The encoder needs about as much computation per crop as SAM's ViT-H, much more than SAM 2's.

OpenCL device

With device=auto, the first GPU is used, else the first CPU device. When several OpenCL platforms expose the same device (for example Mesa's Clover and rusticl drivers for AMD GPUs), the first platform on which the kernels compile is used; platform restricts the choice to platforms whose name contains the given text. Mesa's rusticl only exposes a GPU when enabled in the environment, e.g. RUSTICL_ENABLE=radeonsi. A CPU OpenCL implementation such as PoCL gives the same result, much more slowly.

The weights used take 1739 MiB of device memory (the checkpoint's text encoder, detector and tracker memory are not loaded) and the work buffers 2497 MiB with batch=32; a smaller batch reduces the mask decoder's share. Buffers are kept below the device's maximum allocation size.

Device profile and precision

The profile chooses how buffers are allocated and which matrix product kernels run, to suit the device:

With profile=auto the profile is detected: cpu for a CPU device, igpu for a GPU that reports unified memory, egpu for a GPU whose PCI function Linux marks as removable (when the driver reports its PCI address), dgpu otherwise. The chosen profile and the reason are printed. A profile can be forced to compare them; every profile gives the same results, only the speed differs.

precision=fp16 stores the weight matrices of the linear layers (most of the model's memory) as IEEE half-precision numbers, read with the core OpenCL vload_half functions, so no half-precision extension is needed; the arithmetic stays in single precision. It halves their memory and the bytes read by the matrix products, which matters on devices limited by memory bandwidth (integrated GPUs, CPUs). Rounding the weights to half precision changes the masks slightly; check the result against precision=fp32 (the default) on a sample before relying on it.

Performance

On a CPU (Intel Core i7-1165G7, 4 cores, PoCL), the image encoder took about 1020 s per crop and the mask decoder about 46 s per 64 prompts; with text prompts, the concept detector and its masks took about 75 s per prompt and crop. A GPU is expected to be one to two orders of magnitude faster. The number of crops, and so the run time, grows with the region size divided by tile_size, and with crop_layers.

Choosing the resolution

SAM 3 sees every crop at 1008 x 1008 pixels. Objects should span at least ten or so pixels at that scale: set the region resolution and tile_size accordingly. A smaller tile_size zooms in (crops are upsampled to 1008 pixels), which finds smaller objects at the cost of more crops. Square tiles avoid distorting objects: a crop much longer than wide is squashed to a square.

Text encoder and tokenizer

The text encoder (1.2 GiB of weights) is loaded, run once for all the prompts and released before the image is processed. Its tokenizer is a C port of CLIP's byte-pair encoder, reading the vocabulary file installed with the module (bpe_simple_vocab_16e6.txt.gz, from OpenAI CLIP, MIT licence); prompts must be plain ASCII.

Debugging aids

When the environment variable I_SAM3_OPENCL_DUMP names a directory, the 8-bit crops, the encoder inputs, the image embeddings, the high-resolution features (with text prompts: the detector features and the first prompt's text features) and the list of masks are written there. The scripts tests/reference_compare.py and tests/reference_compare_text.py in the source tree compare them with the PyTorch reference implementation. I_SAM3_OPENCL_PROFILE=1 prints the time spent in each OpenCL kernel. Each line gives the event time and the wall-clock time (some drivers, such as rusticl on Intel iris, report wrong event times), and for the matrix products the achieved GFLOP/s, which shows whether a kernel is limited by compute or by memory bandwidth. I_SAM3_OPENCL_DOWNLOAD_URL replaces the download location (https://huggingface.co/facebook/sam3/resolve/main), for a local mirror.

EXAMPLES

Landsat scene (North Carolina sample dataset)

Segment a true-colour composite of Landsat 7 bands 3, 2 and 1, dropping the background mask (the region has 475 x 527 cells); sam3.pt is taken from, or downloaded into, ~/models/sam/sam3:

g.region raster=lsat7_2002_30 -p
i.group group=lsat_rgb input=lsat7_2002_30,lsat7_2002_20,lsat7_2002_10
i.sam3.opencl group=lsat_rgb output=lsat_segments maxsize=50000
v.db.select lsat_segments | head

Text prompts

Find the buildings and the swimming pools of an orthophoto, keeping detections with a score of at least 0.6:

i.group group=ortho_rgb input=ortho.red,ortho.green,ortho.blue
g.region raster=ortho.red -p
i.sam3.opencl group=ortho_rgb output=found \
    prompt="building,swimming pool" threshold=0.6
v.db.select found columns=cat,prompt,score,area | head

Agricultural parcels

Delineate fields in a 2.5 m VNIR image with bands 1-3 grouped as RGB, keeping only segments of at least 0.1 ha (160 cells) and a label raster alongside the vector map:

i.group group=vnir_rgb input=vnir.1,vnir.2,vnir.3
g.region raster=vnir.1 -p
i.sam3.opencl group=vnir_rgb output=fields raster=fields minsize=160

REFERENCES

Carion, N., Gustafson, L., Hu, Y.-T., Debnath, S., Hu, R., Suris, D., et al., 2025. SAM 3: Segment Anything with Concepts. arXiv:2511.16719

SEE ALSO

i.group, i.sam.opencl, i.sam2.opencl, i.segment, r.to.vect, v.clean, v.generalize

AUTHORS

Yann Chemin

SOURCE CODE

Available at: i.sam3.opencl source code (history)

Accessed: Sunday Nov 15 10:57:11 2026


Main index | Imagery index | Topics index | Keywords index | Graphical index | Full index

© 2003-2026 GRASS Development Team, GRASS 8.5.1dev Reference Manual