Finalize SCADS database for Fall 2020 Eval

[stephenbach]

• [x] Need to update SCADS schema with these indices:

  create index idx_nodes_conceptnet_id on nodes(conceptnet_id);
  create index idx_edges_start_end on edges(start_node, end_node);
  create index idx_edges_end_start on edges(end_node, start_node);

Datasets

• [x] CIFAR 100
  • [x] Installer written and tested against July 2020 API
  • [x] Downloaded to grid
  • [x] Stored in database
• [ ] COCO 2014
  • [ ] Installer written and tested against July 2020 API
  • [x] Downloaded to grid
  • [ ] Stored in database
• [x] DomainNet
  • [x] Installer written to store each sub dataset as a separate data set
  • [x] Tested against July 2020 API
  • [x] Downloaded to grid
  • [x] Stored in database
• [x] ImageNet 1K
  • [x] Installer written and tested against July 2020 API
  • [x] Downloaded to grid
  • [x] Stored in database
• [x] MNIST ¯_(ツ)_/¯
  • [x] Installer written and tested against July 2020 API
  • [x] Downloaded to grid
  • [x] Stored in database
• [ ] VOC2009
  • [ ] Installer written and tested against July 2020 API
  • [x] Downloaded to grid
  • [ ] Stored in database
• [ ] Google Images
  • [ ] Installer written and tested against July 2020 API
  • [x] Downloaded to grid
  • [ ] Stored in database

Once all data is loaded let Steve know so we can upload a final copy to a public google cloud storage.

[elahehraisi]

cifar100 stored

[topwasu]

The tests for object detection datasets should be rewritten after we decide on a labeling strategy. I just updated the checklist.

[stephenbach]

I think to decide on a strategy, we need to try a couple of different things and look at the results. So let's propose an initial strategy. Perhaps:

• any image with only one type of label
  • any bounding box that has some threshold of the total image area, say 25%?

[topwasu]

I think using both of them in combined is a good starting point. After that, we can tune the bounding box threshold and add more rules. @elaheh61 What do you think?