SiaSearch v0.5: Custom Dashboards, Attribute Search & More

Alex Nguyen
SiaSearch (now Scale Nucleus)
3 min readJul 14, 2021

Our product and engineering team has been busy building out new capabilities to make managing visual datasets even easier. We’re excited to share several new product updates so you can spend less time digging through data, and more time on model development.

Check out the latest updates to the SiaSearch platform below.

📊 Build and save custom dashboards

Dashboards help you get a bird’s-eye view of your data. They are an excellent tool for tracking and broadcasting KPIs, as well as tracking new features and experiments.

However, we understand that not every attribute may be relevant to the types of analyses you frequently do. To quickly visualize the data you care about most, you can now build and save custom dashboards on SiaSearch. Easily monitor core metrics and key performance indicators with interactive charts, tables, and more.

Currently, SiaSearch dashboards support vertical / horizontal bar charts, pie charts, map plots, and text blocks. Plots can be resized and spatially moved around the canvas. Once you’re happy with the layout, save dashboards for easy access in the future.

Creating multiple dashboards for different workflows (e.g. geographic spread vs behavioral spread) helps you determine which queries to run next. In addition, you can further interact with plots to filter data in real time. This new functionality allows for further interactive exploration by making it easy to translate plot filters into queries.

🔳 Analyze 2D bounding box data attributes

Bounding boxes are commonly used to identify the right object instances quickly for inspection. This can be based on spatial positioning, such as in the case of traffic lights in the center or prediction confidence. The ability to analyze 2D bounding boxes can uncover potentially interesting data items for annotations, training, or further testing.

Previously, users could only query over object counts, leaving a lot of this analysis to be done either visually or in Python notebooks. Now, when building a query in SiaSearch, you can navigate to ‘2D Object Bounding Boxes’ and select from a variety of dimensions.

With this new feature, you can specify:

  • Spatial positioning using any combination of x_min, x_max, y_min, y_max
  • Confidence threshold between 0 and 1
  • Object class and model

These attributes can be used like the rest of the dimensions and can be accessed under [x] in the hierarchy.

🔎 Attribute search improvements

We are excited to announce new enhancements to our search experience. In addition to attribute types, our search bar now searches over categorical and ordinal states.

Previously, search was restricted to specific attribute names, and not the values or states that they can take. As an example, you could query for “Pedestrian Traffic” generally, but not specifically for scenes with high pedestrian traffic.

Now, using the search bar over the query menu, as soon as you begin a query (e.g. “high” , the closest matching results appear as suggestions:

We’re already at work on a bunch of new functionalities coming very soon. To stay up to date on SiaSearch releases as they are shipped, sign up to receive our newsletter.

Want to explore our new features? Check them out now on the public version of our platform.

This article was originally published at: https://www.siasearch.io/blog/product-updates-june-2021

--

--