Skip to content
A server application used for tagging large batches of images for training-set generation.
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.


imgclass is a server application used for tagging (selecting bounding boxes around objects of interest) large sets of images. Jobs can be assigned to individual workers for tagging.

The variety of possible labels for images in a set is completely customizable. For example:
car, van, bus, SUV, truck, person, motorcycle, bicycle, none
cat, dog, turtle, bird

imgclass is intended to be used for neural network research in the construction of supervised training sets.

Once the complete image set has been labeled, a textfile containing bounding boxes for each image can be downloaded.

User Instructions

Default administrator account:
Default administrator password: password

Default worker account:
Default worker password: password

  1. Upload a large set of images to AWS S3 or other cloud file storage platform. Images must be publicly readable.
  2. Generate a text-file containing one image URL per line.
  3. Login to imgclass as administrator. Upload image URLs textfile and define labels
  4. Assign labeling jobs to workers
  5. Workers log in to imgclass and labels images
  6. Administrator downloads completed training set

Server Provisioning and Deployment

  1. Create AWS EC2 t2.micro Ubuntu 16.04 instance
  2. SSH into server
  3. >> git clone && cd imgclass
  4. Follow instructions in to create a new user
  5. Execute >> ./ on server
  6. Close SSH connection
  7. In local development environment, execute >> cap production deploy
  8. In local development environment, execute >> cap production rails:rake:db:setup

Development Environment Setup

  1. Clone imgclass repo
  2. Install rbenv, ruby, etc (follow script)
  3. Launch development server using

The format for inputing images to be tagged is a textfile with each line containing an image URL, ex:

Technical Background

The imgclass application code contains the following models:

  • ImageLabelSet: The main concept linking a set of images to a set of possible tags, and the results from user tagging.
  • ImageLabel: A user's tagged results from a single image
  • Image: A single image to be tagged, possible by multiple users
  • Job: A group of ImageLabels assigned to a particular worker
  • Label: One of several possible object types that may occur in a given ImageLabelSet
  • User: Either an admin (capable of uploading ImageLabelSets and assigning jobs) or a worker (capable of tagging images).

Images were previously stored directly on the Rails application server but can now be stored on an external server and referenced by URL. This is the recommended choice.

Image tags are represented as JSON-formatted rectangles:


The image-tagging page renders bounding-boxes as SVG. The form transmit a JSON representation of the bounding-box set to the Rails server for storage.

The performance demands on the server should be relatively light-weight using the external URL strategy.

You can’t perform that action at this time.