Foreground segmentation

DeepLabV3+ MobileNetV2 (VOC)

A 23 MB segmentation model for the 21 Pascal VOC classes.

Apache-2.0

Explore model

About this model

DeepLabV3+ paired with a MobileNetV2 backbone. Trained on Pascal VOC for 21 classes: background plus 20 foreground categories like person, car, cat, and dog. The export bakes argmax into the graph and returns a uint8 class-index map at 520x520.

The build is the Qualcomm AI Hub float export, around 23 MB. On M1 Max with CoreML, warm runs land at 37 ms on a 1280x720 image.

What you can build

Cut out a person from a photo with a small, fast model.
Tag pets and vehicles in user uploads at 27 fps on Apple Silicon.
Build a quick foreground mask for downstream effects.
Pre-segment a frame before applying ByteTrack on detected people.

Variants & downloads

Choose the build for your app

Model files are being prepared.

Explore the model’s examples, source links, and integration notes while download variants are added.

Example inference

A multi-person restaurant scene used as the segmentation fixture. — Input

The same scene with foreground people highlighted by class color. — Output

Performance

Results depend on the device, runtime, and test conditions. Each measurement keeps its own context.

M1 Max, macOS 15, ONNX Runtime 1.17

See original catalog hardware notes · CPU

679.5 ms

Warm inference

Catalog reference · variant unspecifiedCold start (ms): 689.04

Reference result from the existing catalog; conversion variant and measurement date were not recorded.

M1 Max, macOS 15, ONNX Runtime 1.17

See original catalog hardware notes · CoreML

36.6 ms

Warm inference

Catalog reference · variant unspecifiedCold start (ms): 97.7

Reference result from the existing catalog; conversion variant and measurement date were not recorded.

RTX 2060, Ubuntu 24.04, ONNX Runtime 1.24.3

See original catalog hardware notes · CUDA

38.2 ms

Warm inference

Catalog reference · variant unspecified

Reference result from the existing catalog; conversion variant and measurement date were not recorded.

Integration examples

Vision SDK integration

cpp

#include "imaged/sdk.hpp"
#include "imaged/families/semantic_segmentation/semantic_segmentation_family.hpp"

imaged::SDK::AI ai;
auto& config = ai.getOptions();
config.set("semantic_segmentation.model", "weights/deeplabv3_plus_mobilenet.onnx");
config.set("semantic_segmentation.normalize", "scale255");

imaged::SDK::SemanticSegmentationFamily family;
family.load(config);

imaged::SDK::Image image;
image.load("scene.png");

auto result = std::get<imaged::SDK::SegmentationResult>(family.run({&image}));
// result.mask is W*H bytes, each pixel a VOC class index 0..20