#26091·ultralytics

Generic Detection-to-OCR Solution

Author: ojedavdCreated Sep 8, 2026Updated Sep 8, 2026
Labelsenhancementquestionsolutions

Search before asking

  • I have searched the Ultralytics issues and found no similar feature requests.

Description

I would like to propose a generic OCR workflow under ultralytics/solutions/ that combines YOLO-based region detection with a user-provided OCR recognizer.

I found the previous OCR-related discussion in #5716. This proposal is intentionally narrower and does not aim to add OCR capabilities to YOLO itself or introduce a built-in OCR model into Ultralytics.

Instead, the proposed solution would use YOLO only for region detection and provide the orchestration required to crop detected regions, pass them to an external OCR recognizer, and associate the recognized text with the corresponding detections.

Motivation

Ultralytics already provides high-level solutions for common computer vision workflows such as object counting, parking management, speed estimation, and region-based analysis.

A generic OCR solution could extend this concept to use cases where text must be recognized inside detected regions, for example:

  • License plates
  • Shipping labels
  • Industrial labels
  • Serial numbers
  • Signs
  • Product labels
  • Meter displays

The solution should not be tied to any particular OCR model or application domain.

Proposed Pipeline

Image
  ↓
YOLO model provided by the user
  ↓
Detected regions
  ↓
Crop each bounding box
  ↓
User-provided OCR recognizer
  ↓
Recognized text
  ↓
Annotation + structured results

The YOLO model would be responsible only for locating the regions that should be processed.

For example, the user's model could detect:

text
license_plate
shipping_label
serial_number

The OCR component would then recognize the text contained inside each detected region.

Before implementing this, I would appreciate feedback on:

  • Would this type of detection + OCR orchestration fit the scope of ultralytics/solutions/?
  • Would accepting a user-provided OCR callable/backend be preferable to integrating a specific OCR library?

I would be happy to work on an implementation and tests if this fits the project direction.

Use case

No response

Additional

No response

Are you willing to submit a PR?

  • Yes I'd like to help by submitting a PR!