v2.4 removed outdated configuration options. Please see the list of supported Configuration Options.
enable_corruptness_checkhas been removed. The corruption check will always be done.
remove_exact_duplicateshas been removed. Duplicates will always be removed.
upload_datasethas been removed. Please use a datasource instead. For setting up a datasource and using it, see:
datapool.keep_historyhave been removed.
use_video_reader_backendhas been removed.
use_redirected_read_urlhas been removed. The feature is automatically enabled.
use_frame_count_metadatahas been removed. If the frame count metadata of videos is wrong, the run will break, thus it will never use them and instead compute the frame count by actually iterating over the video frames.
scorer_confighave been removed. Please see Customize a Selection instead.
append_weak_labelshas been removed. Please see Work with Metadata.
pretagging_confighas been removed. If you want to balance based on predictions, head to Customize a Selection.
active_learning_score_column_namehave been removed.
- Using Custom Embeddings is still possible, however, it will only be used for providing embeddings.
- Samples that are already selected are handled automatically by the Datapool feature.
- Only considering a subset of the samples in the input can be done using the Relevant Filenames feature.
- Active learning can be done using the scores as input. If you want to provide your own scores, you can provide them as Metadata.
n_example_imageshas been removed. The report will always show 6 example images.
show_video_sampling_timelinehas been removed. The video density plots are always included in the report.
pretagging_debughas been removed. Examples of pretagging are always shown.
memory_requirement_in_GBhas been removed,
- The configuration option
object_levelis deprecated in
v2.4and will be removed in
v2.6. Please move to the new Object Diversity workflow instead.
Updated 6 months ago