slide_cbir_47

A pyqt5 GUI for CBIR in whole-slide images with Openslide.

Project depends on:

• cbir_47, which depends on
  • keras
  • skimage
  • sklearn
  • quantization_47
  • common_utils_47
• slide_viewer_47
• slide_list_view_47

Implemented CBIR-technique:

Precompute stage. For each db-slide (slide from collection of slides you want to search):

• slice each db-slide image into tiles
• compute descriptor of some type for each tile

Search stage. For a given query-tile (tile from some slide image):

• compute descriptor of some type (corresponding to types selected in precompute stage)
• exhaustively compare each descriptor of each db-slide with descriptor of query-tile
• visualize distances between db-tiles and query-tile

Use:

• set up configuration-file (configuring descriptor types, tile size, path to db-slides, path to descriptors to be computed)
• use download_utils to load some whole-slide images if you havent
• generate json-models for future computation
• run descriptors computation (time-consuming)(very time-consuming if you selected vgg16 descriptor type)(one slide image~300mb might take>100 seconds)
• configure ui-specific options
• run gui app
• select collection of json-models with proper db-slides to search in (left pane)
• select tile in query-slide viewer (right pane)
• select in menu "actions"->"search in selected db_models"
• search results (represented as intensities proportional to distances between tiles) will be populated (bottom pane)

Notes:

• for any db-slide or result-slide there is an extended view mode (activated by doubleclick) where you can zoom it and toggle grid visibility.
• descriptors are stored in hdf5 files.

Examples of whole-slide images can be downloaded from openslide-testdata page.

Screenshot of slide_cbir_47: