July 18, 2021

This release is packed with updates that improve your experience with datasets and neural networks.

Explore feature

Publish dataset

You can make a dataset public. Anyone with the link to a published dataset can see the dataset and run a query.

To publish a dataset:

  1. In the top right corner of the Explore page, select Publish.
  2. Choose Publish.
  3. Copy the dataset’s link to share it.

When a dataset is published, a view count is displayed in the top bar. The view count is available in the editor and the viewer mode.

The view count represents the number of users who have opened the public dataset’s link. Opening the link from a new tab or a new window is counted as a view while refreshing the page is not considered a new view.

Unpublish dataset

  1. In the top right corner of the Explore page, select Published.
  2. Choose Stop publishing.

🚧

When a dataset is unpublished then published again, the previous URL will be invalid. So you will have to share the new URL.

📘

Other features and improvements

  • The Explore feature now works with images that are linked from external storages.
  • If you want to show the images that contain unclassified instances, check the Unclassified instances box.
  • To get the download link for a dataset’s JSON file, select the link icon in the top right corner. The link is valid for two days.
  • You can now see the instances count in the left panel.

Neural network

Train/Test data

You now have more control over defining the train/test split. This will help you get reproducible training experiments and compare multiple models against the same test data. Learn more.

Model training function

We changed the model training function to:

sa.run_training(
  model_name = "Cityscapes Segmentation Model", 
  model_description = "Model Finetuned on Cityscapes", 
  task = "Instance Segmentation for Vector Projects",
  base_model = "Instance Segmentation (trained on COCO)", 
  train_data = ["Cityscapes/Batch 1", "Cityscapes/Batch 2"], 
  test_data = ["Cityscapes/Test"], 
  hyperparameters = {"base_lr": 0.02, "images_per_batch": 8},
  log = True)

What do you think of the updates? We’d love to hear your feedback.

We have a blog. Follow us to stay updated.