Classifier vs. Segmenter

This is the most common question when creating a recipe. Here's a clear framework.

Classifier

A classifier looks at each ROI crop and assigns it to a category: pass, fail, present, absent, scratched, good, or whatever classes you define.

Think of it like sorting cards into piles. The AI looks at the whole crop and says "this goes in the pass pile" or "this goes in the fail pile."

Use a classifier when:

• You need pass/fail or multi-class decisions
• Your ROIs are small enough to clearly show the defect
• Labeling speed matters (classifiers are much faster to label)
• You're checking for presence/absence of a feature
• Examples: screw present/absent, connector seated/unseated, label correct/wrong

Need training images?

Use the Defect Creator Studio to generate photorealistic synthetic defect images for either approach — describe the defect in plain English and get training-ready images in seconds.

Training Modes

Classification offers two distinct training approaches:

Fast Classifier

• Purpose: Quick setup and testing
• Use Case: Rapid prototyping and proof-of-concept
• Speed: Faster training time
• Accuracy: Lower precision
• Production Ready: Not recommended for production

Accurate/Production Classifier

• Purpose: Production-ready inspections
• Use Case: Final deployment and manufacturing
• Speed: Longer training time
• Accuracy: Higher precision
• Production Ready: Main training model for best results

Classification Examples from Production

Loose Nut Detection (OEM Assembly)

Task: Identify loose nuts in mechanical assemblies

Setup:

• ROI: Rectangle around nut area
• Classes:
  • Good - Properly tightened nut (Pass 100%)
  • Defect - Loose nut (Fail 98%)

Result: Binary decision - nut is properly seated or loose

Shaft Seating Verification (OEM Assembly)

Task: Identify partially seated shafts

Setup:

• ROI: Rectangle covering shaft mounting area
• Classes:
  • Good - Fully seated shaft
  • Defect - Partially seated shaft
  • Defect - Fully out shaft

Result: Component gets categorized as properly installed or defective

Radiator Pin Inspection (T1 Supplier Fabrication)

Task: Check for dents in individual radiator pins

Setup:

• ROI: Rectangle covering radiator pin array
• Classes:
  • Good - Undamaged pin (Straight)
  • Defect - Bent/Damaged pins (Bent)

Result: Classifier differentiates a good and a bad pin

Segmenter

A segmenter examines each ROI crop at the pixel level and identifies exactly which pixels belong to a defect.

Think of it like highlighting with a marker. The AI circles exactly where the defect is, not just whether one exists.

Use a segmenter when:

• You need pixel-level defect localization (where exactly is the scratch?)
• You need to measure defect size or area
• Your ROIs need to be large but you're looking for small defects
• You need to count multiple defects in one region
• Examples: surface scratches, stains, cracks, weld quality, dimensional measurements

Training Mode

Segmentation uses a single training mode:

Accurate/Production Only

• Purpose: Production-ready pixel-level analysis
• Training: Single high-accuracy mode
• Speed: Longer training time for precise pixel classification
• Accuracy: High precision for detailed defect mapping
• Production Ready: Optimized for manufacturing environments

Segmentation Examples from Production

Surface Defect Detection (T1 Supplier Assembly)

Task: Identify glue spill

Setup:

• ROI: Rectangle around valve surface
• Pixel Classes:
  • Glue (Yellow pixels)

Result: Yellow highlights show exactly what the AI model detected as defects, with raw images shown for reference

Gap Size Measurement (T1 Supplier Assembly)

Task: Verify that the gap of the inspection is the right size

Setup:

• ROI: Rectangle around gap area
• Measurement Logic: Pass condition based on pixel area values
• Classes:
  • Blue pixel teaching what distance looks like
  • Good - Area < 100 pixels (Pass)
  • Defect - Area > 100 pixels (Fail)

Result: Precise measurement of gap size with exact pixel count

Paint Splatter Detection

Task: Identify paint splatter on a surface

Setup:

• ROI: Rectangle around the inspection region
• Pixel Classes:
  • Yellow pixel to teach what paint looks like

Result: Pixel-level detection of paint splatter

Inspecting Multiple Features Simultaneously

A segmenter can inspect multiple features at once by defining separate pixel classes. For example, you can train a single segmenter to detect scratches (class 1), dents (class 2), and stains (class 3) in the same ROI. Each class is trained independently by painting pixels of that defect type with a different label. The segmenter identifies all defect types in a single pass, reporting which pixels belong to which class.

This is different from classification, where the entire ROI gets a single label. With segmentation, multiple defect types can coexist in the same image and each gets its own pixel-level mask.

Comparison

Aspect	Classifier	Segmenter
What it tells you	Category (pass/fail/etc.)	Exact pixels of the defect
Labeling effort	Low: select a class per image	High: paint defect pixels
Data needed	3-5 images per class to start	Similar, but labeling takes longer
ROI size	Works best with small ROIs	Handles larger ROIs better
Best for	Presence/absence, category decisions	Scratches, stains, measurements, pixel-level detection

Cycle Time Performance

Model Type	Inspection Speed	Best Use Case
Segmentation	Fastest	High-speed production lines
Classification (Accurate)	Slower	Quality-critical inspections
Classification (Fast)	Medium	Testing and prototyping

note

Segmentation models provide the fastest inspection cycle times, making them ideal for high-throughput production environments.

When to Choose Each Model

Choose Classification When:

• You need simple good/bad categorization
• The entire ROI can be labeled as one of several classes
• Overall condition assessment is sufficient
• Fixed defect locations are expected
• You want faster labeling during setup

Best for: Go/No-Go decisions, component presence/absence, overall quality assessment

Choose Segmentation When:

• Defects can appear anywhere in the inspection area
• You need to measure size or area of defects
• Multiple defect types may exist in one image
• Precise location mapping is required
• Fastest cycle times are critical for production

Best for: Surface defects, gap measurements, coverage verification, variable-location flaws

When in doubt, start with a classifier

Classifiers are faster to set up, need fewer training images, and work well for most pass/fail inspections. You can always switch to a segmenter later if you need pixel-level detail. Starting simple lets you validate your setup quickly.

Segmenters require more setup time

Segmenters need pixel-level annotation during labeling, which takes significantly longer than selecting a class for a classifier. Budget extra time for the initial setup and labeling compared to classifiers, especially if you have many training images.

The 512x512 pixel processing limit

Both classifiers and segmenters process ROI crops at a maximum resolution of 512x512 pixels. Any ROI larger than this gets downscaled before the AI processes it, which means small defects may become invisible. Keep your ROIs as small as possible to preserve detail, or use a segmenter with a larger ROI if you need to cover a wide area while still detecting fine defects.

The overlap zone

There's genuine overlap. Many inspections could work with either approach. If you're not sure:

1. Start with a classifier. It's simpler to set up and faster to label
2. If you find the classifier isn't precise enough (can't detect small defects in large areas, or you need measurement), switch to a segmenter
3. Ask the AI Assistant at tools.overview.ai. Describe your application and it will recommend the best approach

Training Mode Summary

Training Mode	Model Type	Production Ready	Notes
Fast Classifier	Classification	No	Quick setup for testing and prototyping
Accurate Classifier	Classification	Yes	Production-ready with higher precision
Accurate Segmenter	Segmentation	Yes	Single high-precision mode optimized for production

OV20i vs OV80i

• OV20i: Choose one per recipe, either a classifier or a segmenter
• OV80i: Multi-model support. Run a classifier and a segmenter simultaneously on the same capture within a single recipe. For example, use a classifier ROI to check component presence (pass/fail) while a segmenter ROI detects surface defects on the same part. Both models run on every trigger and produce independent results.