Model reference — Pictures
Segmentation marks out the exact shape of things in an image rather than a rectangle around them, labelling every pixel as belonging to one object or region.
The exact shape is what allows measurement. A box tells you a crack is present; a mask tells you it is 4 millimetres wide and 60 long. Area, coverage and volume all follow from masks: the proportion of a field affected by disease, the area of corrosion on a hull, the fraction of a shelf that is empty. Masks are also what make clean image editing possible — removing a background, blurring only the people, replacing one surface — and they are the ground truth from which many other models are trained.
Models in this group take a single image as input: a photograph, a scan or a screenshot. Each specification table gives three hardware tiers — Minimum, the smallest setup on which the model runs correctly; Medium, the usual production configuration; and High, a configuration sized for peak volume. The rate is the number of sample inputs processed per hour on that hardware. Use these rates for initial sizing. Before production, benchmark your own data to validate accuracy, latency, throughput and cost. The sample input here is one photograph at about 2 megapixels.
Vendor: Meta
What it does: outlines any object you point at or box, with no training on your objects at all, and returns a precise mask. The most generally useful segmentation model we deploy.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3090 | RTX 4090 | A100 80 GB |
| VRAM | 12 GB | 24 GB | 80 GB |
| vCPUs | 8 | 12 | 24 |
| RAM | 32 GB | 48 GB | 96 GB |
| Server | 1× RTX 3090 24 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (images/hour) | ≈ 3,000 | ≈ 12,000 | ≈ 30,000 |
Vendor: Meta
What it does: labels every pixel in the image against a fixed set of categories, which is what you want when the whole scene must be accounted for rather than one object outlined.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3090 | RTX 4090 | A100 80 GB |
| VRAM | 12 GB | 24 GB | 80 GB |
| vCPUs | 8 | 12 | 24 |
| RAM | 32 GB | 48 GB | 96 GB |
| Server | 1× RTX 3090 24 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (images/hour) | ≈ 4,000 | ≈ 16,000 | ≈ 42,000 |
Vendor: NVIDIA
What it does: a smaller and faster whole-scene segmenter, easy to fit to your own regions and light enough to run on modest hardware or a camera device.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 8 GB | 24 GB | 80 GB |
| vCPUs | 6 | 12 | 24 |
| RAM | 16 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (images/hour) | ≈ 12,000 | ≈ 45,000 | ≈ 130,000 |
Vendor: Ultralytics
What it does: detects objects and outlines them in one pass, at close to detection speed. The practical choice when masks are needed on every frame of a production line rather than on selected images.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 6 GB | 24 GB | 80 GB |
| vCPUs | 4 | 8 | 24 |
| RAM | 16 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (images/hour) | ≈ 25,000 | ≈ 100,000 | ≈ 280,000 |
Vendor: IDEA Research
What it does: finds the object from a written phrase so that a segmentation model can outline it, which together allow "outline every cracked tile" with no training data.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3090 | RTX 4090 | A100 80 GB |
| VRAM | 12 GB | 24 GB | 80 GB |
| vCPUs | 8 | 12 | 24 |
| RAM | 32 GB | 48 GB | 96 GB |
| Server | 1× RTX 3090 24 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (images/hour) | ≈ 2,000 | ≈ 8,000 | ≈ 20,000 |
Vendor: Meta
What it does: produces general-purpose image features from which a small segmentation head can be trained on very few labelled examples — useful when labelling is expensive.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 8 GB | 24 GB | 80 GB |
| vCPUs | 6 | 12 | 24 |
| RAM | 16 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (images/hour) | ≈ 8,000 | ≈ 30,000 | ≈ 85,000 |
The right model depends on whether you need every region labelled or only certain objects outlined, and on the precision your measurements require. Our consultants review your images and what is being measured, then recommend a model and the resolution it should run at.
At the start of a project we may run a short proof of concept on a sample of your own data, measuring the accuracy and the throughput the model actually achieves on your material. That replaces the estimates on this page with real figures, so the cost and the schedule for the full engagement are known before it is committed.
Share a representative sample, expected volume, latency target and deployment location for Image segmentation. We benchmark the shortlisted models, recommend the lowest-cost GPU configuration that meets the target, and scale it from pilot capacity to a dedicated production cluster.