Skip to content

supervisely-ecosystem/dev-smart-tool-batched

Repository files navigation

Batched Smart Tool

OverviewUsageHow to run

GitHub release (latest SemVer) views runs

Overview

Application allows you to label objects with Smart Tool using batch way.

Application key points:

  • Multiclass annotation
  • Input data ordering
  • Saves annotation progress automatically
  • Apply model to Objects (Rectangles) / whole Images
  • Mark broken Objects / Images
  • Linked green / red points between cards
  • Flexible settings
  • Doesn't change the Input Project

Usage

📋 Content:

Application Startup

  1. On startup application will suggest input data order.
    For comfortable and efficient labeling we recommend to use sorted bboxes order.

  1. Application saves annotation progress automatically.
    If you launch application on same project a second time — it will suggest you to continue the labeling process from the paused point.

Controls with Shortcuts

Key Description
Left Mouse Button Place a green point
Left Mouse Button + Shift Place a red point
Left Mouse Button + Ctrl Remove point
Scroll Wheel Zoom an image in and out
Right Mouse Button + Move Mouse Move an image
Shift + E Link All Cells
Shift + Q Unlink All Cells
Shift + A Assign Points Automatically
Shift + D Update Unupdated Cells
Shift + C Clean All Linked Cells

General Usage Scenario

  1. Assign Base (Shift + A) points to all linked cells

  1. Update Masks (Shift + D) on all unupdated (orange) cells

  1. Unlink All (Shift + Q) cells

  1. Easily place green points to label-interested area and red points to label-not-interested area to correct local mistakes.

  1. When you satisfied with results, click Next Batch button to load next images

Cells Functionality

Batched Smart Tool consists of cells with objects to label.
Each cell has core functionality:

  1. Linked Cell — all points (green / red) on linked cells will synchronise.
  2. Mark as Unlabeled — object will be marked with _not_labeled_by_BTC tag in output project.
  3. Mark as Labeled — mark object as labeled (optional option).
  4. Show in Input Project — open object in the Classic Labeling Interface.

Preferences Panel

Preferences Panel allows you to customize labeling interface for your needs.

Grid Cells Preferences allows you to adjust number, size, padding and masks opacity parameters for each cell in the grid.

Model Preferences allows you to connect to served Smart Tool model, and select labeling mode.

Apply model to Preference allows you to choose input data for labeling.
You can select objects (Rectangles) from prepared project or label whole images.

ℹ️ Multiclass labeling available

Control Panel

Control Panel allows you control labeling process.

Link all (Shift + E) — links all cells, all points (green / red) on linked cells will synchronise.
Unlink all (Shift + Q) — unlinks all cells.

Assign Base (Shift + A) — assign 8 red points to corners and 1 green point to center of image.
Clean Up (Shift + C) — clean up data from all linked cells.

Update Masks (Shift + D) — updates masks on all unupdated (orange) cells.
Next Batch — uploads labeled data from cells to output project and load new data to label.

How to run

  1. Prepare Images Project

Option 1: with roughly assigned Rectangles

Option 2: with plain images

  1. Launch RITM interactive segmentation Smart Tool

  1. Add Batched Smart Tool to your Team

  1. Launch from context menu of prepared project