FSS-1000: A 1000-Class Dataset for Few-Shot Segmentation 

This zip file contains 5 PDF files:

1. hardcase_mining.pdf: A supplement to Figure 11 in the main paper. It takes "Eiffel Tower" as an example, showing the effect of mining hard cases and adjusting support set accordingly. Ten query examples are picked for zoom-in demonstration.

2. multiple_instances.pdf: Examples of unseen classes where multiple instances occur in a query image, showing the potential benefit of few-shot segmentation model to applications such as novel object instance segmentation and counting.

3. FSS-1000_classes.pdf: Full hierarchy of FSS-1000 dataset.

4. network_architecture.pdf: It contains a table illustrating per-layer setup of our network. ReLU activations following every convolution layer are omitted. The last layer is followed by Sigmoid activation.        

5. more_experiments.pdf: It contains the results of more experiments, including models trained on fsPASCAL.