---
title: "Utility vision: see, cut, and sharpen"
lesson_id: "04"
---

# Utility vision: see, cut, and sharpen

- **Lesson ID:** 04
- **Goal:** Map utility vision task labels to classification, detection, segmentation, depth, and super-resolution presets.
- **Human lesson:** [04-see-and-cut.html](04-see-and-cut.html)

## Prerequisites

- Lesson 01 and a simple picture of an image going to an output.
- The trained ClothingCNN path in [the forward course](../pytorch-core-ai-xcode/).

## Inputs, outputs, and artifacts

- **Inputs:** An existing image and a requested visual output.
- **Outputs:** A task-first chooser for labels, boxes, cut-outs, depth, and sharper images.
- **Artifacts:** Progressive `.flow` diagrams and the ClothingCNN versus `pvt-v2-b0` comparison.

## Agent build steps

1. Explain `utility` as a toolbox whose task field is the label to teach.
2. Map `pvt-v2-b0` to image classification and place trained ClothingCNN beside it as the same broad job class.
3. Map `yolos-base` and `yolos-tiny` to object detection and boxes.
4. Map `efficient-sam-vitt` and `sam3` to segmentation and precise object cut-outs. Say `sam3` can use text or visual prompts.
5. Map `depth-anything-3-small` to depth and state that the registry marks it macOS only.
6. Map `edsr-x2` to super-resolution for a sharper or larger image.
7. Link the architectures, forward, and From the metal to the model.

## Constraints

Do not turn a utility into a general assistant or invent extra model names. No math, CUDA, quizzes, Mermaid, or em dashes.

## Key concepts

- Utility vision tasks are `classification`, `detection`, `segmentation`, `depth`, and `super-resolution`.
- `pvt-v2-b0` is the catalog classification preset. Same broad job as trained ClothingCNN.
- `yolos-base` and `yolos-tiny` return boxes. `efficient-sam-vitt` and `sam3` return cut-outs.
- `depth-anything-3-small` is depth and macOS only. `edsr-x2` is super-resolution.

## Takeaways

- You can pick a utility vision task by the output the app needs.
- You can map labels, boxes, and masks to the named presets in this lesson.
- You can check `depth-anything-3-small` is macOS only before you pick it.

## Acceptance checks

- Classification, detection, segmentation, depth, and super-resolution are distinct.
- All supplied vision utility names are present exactly.
- ClothingCNN is described as trained while `pvt-v2-b0` is a registry preset.
- SAM's cut-out job and `sam3` text or visual prompts are explained.
- macOS-only depth coverage is stated.
- Progressive `.flow` diagrams, prior-course links, and the Agent brief link are present.