Get AsyncJob by ID

• Add NucleusClient.get_job to retrieve AsyncJobs by job ID

Add AP and mAP to Polygon Metrics

What's Changed

• Add average precision and mean average precision to polygon metrics

Remove shapely dependency

What's Changed

• Add dataset.delete_scene
• Remove shapely dependency

Fix python 3.6 compatibility

Fixed dependencies to be fully compatible with python 3.6

Add Metrics

• Adds IOU, precision, and recall for box/polygon annotations and predictions

List Dataset Scenes

• Add Dataset.scenes property that fetches the Scale-generated ID, reference ID, type, and metadata of all scenes in the Dataset.

Export raw Slice items

• Add Slice.export_raw_items(), which exports accessible (signed) URLs pointing to the raw item on Scale's servers

Categorization Improvements

• Updatable category taxonomies
• Improved error messages for categorization

QoL Improvements for Slices

• Add Slice.name
• Add Slice.items
• Speed up Slice.info by not fetching Slice items
• Deprecate NucleusClient.slice_info in favor of Slice.info

Differentiation between sequence and item-level datasets

The changes in this PR introduce a new category of Nucleus Dataset: the scene dataset. The purpose of the is_scene parameter to the NucleusDataset constructor is to differentiate between datasets composed of sequences and datasets composed of individual items. Scenes can only be uploaded to datasets created by setting the is_scene=True, and vice versa, items can only be added to datasets where is_scene=False