Image only collections

To create a collection in Zegami, it has always been necessary to supply a tabular data file (csv or xlsx) providing the ‘row’ data, which includes references to images and other metadata to accompany each one.

This makes perfect sense for many use cases where there is an intrinsic data counterpart to each image. For example data about the treatment and capture time of images in a Plant Phenomics experiment

In other situations however, users come to us with nothing more than a large set of images. A typical requirement is to cluster the images by similarity and take advantage of Zegami‚Äôs highly efficient tagging workflow to label the set. In these situations, it’s an unwelcome overhead to generate a boilerplate csv file just to make the collection. Adding more images also entails uploading an updated csv file. 

To address this issue, we’ve added the ability to create Zegami collections with only images. It’s never been easier to create a Zegami collection and should take only minutes to start gaining insights into a set of image files.

Step 1 

Create a Deepzoom collection, setting some basic information. Only a title is needed. 

Step 2

Drag image files onto the uploader.

Step 3

Await processing.

Step 4

View the results.

The resulting collection can be found here. This view shows the flags arranged in a graph according to the file size.

We include some data extracted from the images themselves such as dimensions and the original filename. These can be used for filtering or sorting just like any data that would be provided via csv. 

Collections made this way can be updated easily. More images can be added at any time. The similarity data will be regenerated for each batch that gets uploaded. Furthermore, a data file can be added at any time, and the collection will then seamlessly convert to the more familiar data-based Zegami collection.

A typical workflow now might consist of using the scatter plot filter to isolate groups of similar images, and add tags via the tag and select workflow. The results can then be exported to a variety of formats to be used for ML training or other applications where labelling is required.