fastdup is a powerful free tool designed to rapidly extract valuable insights from your image & video datasets. Assisting you to increase your dataset images & labels quality and reduce your data operations costs at an unparalleled scale.
This PR fixes a few bugs and improves the overall flow of the gallery captioning features in fastdup:
models are now loaded through the pipeline instead of the current method
model definitions are written as a list of available models, and gallery functions check to see whether the label_col argument is in the list of models
a bug in the visualizations file was fixed, wherein the captioning call was ignored if labels existed in the annotations. Now, if a captioning call is made, the captioning will be forced even if labels already exist in the annotations.
-unnecessary code repetitions are deleted and condensed into callable functions
the exception message is improved for clarity and professionalism
Note: this change only affects captioning applied to galleries. No capabilities for captioning the entire dataset are included in this change.
dbickson
commented
10 months ago
This PR fixes a few bugs and improves the overall flow of the gallery captioning features in fastdup:
Note: this change only affects captioning applied to galleries. No capabilities for captioning the entire dataset are included in this change.