Changelog

Update 2020-01-27

Added:

  • Object name is shown next to the bounding box or polygon
  • Image labels are now downloaded as a .zip file and works faster
  • Now we report both label-wise and image-wise accuracy in multi-label classification
  • Possibility to upload classification labels from .json file (rather than .csv)
  • Possibility to rotate image counter-clockwise

Update 2019-11-20

Added:

  • Possibility to download similarity search results
  • Improved object detection/segmentation labeling interface:
    • Bounding boxes and polygons that are being edited are highlighted in an image and in the list on the right panel
    • Active vertices are highlighted
    • A new vertex which is about to be added on an edge is visualized
    • After finishing labeling a bouding box or a polygon its name field is automatically activated
    • Occluded objects are marked by a dashed line
    • New shortcuts for labeling (‘O’ – mark object as occluded, ‘V’ turn on/off visibility of an object)
  • Added a search field by image filename
  • A new button to select all images in the current page
  • Five new video tutorials (training a model on one class, similarity search, labeling images for object detection, labeling images for segmentation, labeling settings)

Changed:

  • Removed “Move” and “Edit” modes in image labeling (everything is done in a single mode now)
  • Restructured similarity search results window
  • Slightly changed REST api output format
  • User projects with duplicated names were renamed
  • Download labels function now downloads classification labels in .csv file in addition to all types of labels in .json file
  • When a user exits labeling, the current page stays where the last image was and the last image is highlighted

Update 2019-09-20

Added:

  • New shortcuts for labeling (‘Z’ to fit image to screen, ‘Ctrl or Shift + mouse wheel’ to scroll image left/right or up/down)

Changed:

  • A minimum value of optimized score threshold for multi-label classification is now 5%
  • Downloaded image labels for object detection, segmentation and classification are now all in one .json file
  • Reordered context menu of right mouse button

Update 2019-08-30

Added:

  • Functionality to perform image similarity search via web interface and REST api
  • Possibility to see how many images will be allocated to train and validation sets
  • Functionality to rotate image
  • Functionality to sort images by name or upload time
  • Possibility to select images from several different pages
  • New shortcuts: press ctrl+1, ctrl+2, etc. to select object label
  • New shortcut: delete images with [delete] key
  • Possibility to pre-process images for classification and similarity search during image upload

Changed:

  • Restructured project selection and editing menu
  • Removed boundary groups from object detection labeling (now each bounding box is stand-alone)

Update 2019-06-26

Added:

  • Functionality to upload object detection and segmentation labels from a .json file
  • Functionality to select and move polygons by mouse click during labeling
  • Full screen mode for image labeling
  • Icons to mark images labeled for object detection and segmentation
  • Possibility to exclude labels from training in the training screen
  • Possibility for changing line width for labeling
  • Functionality to filter images by labels using “and” and “or” filters
  • Functionality to upload images by selecting a whole folder

Changed:

  • Removed “save” button from settings menu (now settings are saved automatically)
  • Now images in “show predictions” screen are shown in original aspect ratio
  • Restructured upper menu

Updates 2018-11-18 – 2019-06-26

Added:

  • Possibility to upload classification labels as a .csv file
  • Possibility to download classification predictions as a .csv file, .json file or images grouped into folders by predicted labeling
  • Download object detection and segmentation labels as a .json file
  • Upload images as a .zip file with a possibility to label them based on folder structure
  • Possibility to select which exact images will be used for validation
  • Possibility to set “default” label on all selected images
  • Possibility to have a fixed bounding box ratio
  • Possibility to have a fixed number of points in a polygon
  • Possibility to skip images that are already labeled
  • Possibility to switch between images using arrows or “A” and “D” shortcuts

Changed:

  • Labels without the minimum required number of images are automatically excluded from training
  • Improved bounding box and segmentation labeling tools

SentiSight.ai first release 2018-11-18

Added:

  • Functionality to train single-label and multi-label classification models
  • Functionality to label images for classification, object detection and segmentation
  • Functionality to view predictions on train and validation sets and analyze model’s performance
  • Functionality to filter images by type and labels
  • REST api service