Release-1.1

Release 1.1.95 (Bugfix release) - January 10, 2023

1. Fix for an issue where an incorrect full-resolution image was shown for drawing bounding boxes for specifying similarity search requests. This was happening for cases when a saved search from a different job was used.
2. Fix for an issue where analyze job statistics computation was incorrect in certain cases.
3. Fix for an issue where analyze job submission failed if the class column was not of string type.

Release 1.1.91 (Maintenance release) - December 14, 2022

1. Patch similarity search match score calculation improvements.
2. Support AWS type of secret for Athena external catalog integration.
3. Fix for an issue where bounding boxes were not displayed on thumbnails within a resultset.
4. Fix for an issue where patch search with search space tunables failed with an error.

Release 1.1.87 (Bugfix release) - November 30, 2022

1. The account activation email was rendered incorrectly on some email clients on macOS.
2. In certain corner cases, the plot view blinked continuously when hovered over a point.
3. Sometimes, the patch search grid did not render correctly on the Safari browser.

Release 1.1.80 - November 23, 2022

New features

1. Import public datasets: Import one or more public datasets in minutes to quickly start exploring various Data Explorer features.
   
2. Enhanced patch search: Run a search over a smartly sampled larger subset across search space to find the 'needle in a haystack.'
   Quick Search: Searches over a sampled set of images over search space and is suitable for initial tuning of the search set.
   Search: Searches over the larger smartly sampled search space.
3. Draw a bounding box on full-resolution images: When the region of interest is small, it can be hard to express this region of interest on a small thumbnail. Now, you can draw the bounding box for patch search on a full-resolution image in addition to the thumbnail.
   
4.  Show matching regions in patch search: Patch search results show a heatmap of matching areas providing insights into why a particular image/frame is part of the search results.
   
5. Filter bounding boxes displayed: Analyze object detection job supports bounding boxes with class labels and confidence scores. When the number of ground truths or predictions exceeds a handful, it may be hard to correlate which ground truth/prediction matches which bounding box on the thumbnail. With this feature, you can turn on/off specific bounding boxes of interest on the thumbnail and full-resolution images.
   
   

Minor enhancements

1. Enhanced job visualization experience by allowing job submission directly from the catalog page.
2. Four thumbnail size options - S, M, L, and XL in display settings.
3. Improved error messages on the UI.
4. User interface (UI) enhancements.
   1. Add a zoom-in/zoom-out button on the plot view.
   2. Show effective roles on the 'Users' page by combining individually assigned roles and roles inherited due to group membership.
   3. Show the thumbnail until full resolution image gets loaded.
5. Enhanced catalog filtering with starts-with, ends-with, contains, and other operators on a text type of catalog column.
6. Import externally produced features for video type of dataset through a CSV file.
7. Support delete catalog table operation for tables with imported CSV catalog contents.
8. Support updating of secrets for an external catalog. This is targeted to support use cases where organization policies may enforce periodic rotation of credentials.
9. Support deletion of a resultset.

Bug fixes

1. The object detection heatmaps showed values inconsistent with the number of images/frames qualifying under the particular heatmap cell.
2. Fix for a Compare type of job creation failure on WMTS-enabled (satellite imagery in Web Map Tile Service format) datasets.
3. Fix for failure when adectl import to import entries into catalog was invoked with input file having 300K rows.
4. Fix for display issue when adectl setup is run on a terminal with lesser width.
5. In certain corner cases, when the number of clusters is increased/decreased, the legend colors for clusters are inconsistent with those used for points in the plot view.