search

xmed-lab / GraphEcho

ICCV 2023, "GraphEcho: Graph-Driven Unsupervised Domain Adaptation for Echocardiogram Video Segmentation"
MIT License
33 stars 3 forks source link

CardiacUDA dataset #5

Open OrientKing opened 2 months ago

OrientKing commented 2 months ago

When reading this excellent paper, I have the following questions about the "CardiacUDA" dataset. I am looking forward to your reply. Thank you very much.

  1. Video Number：The data sets obtained from a given site https://www.kaggle.com/datasets/xiaoweixumedicalai/cardiacudc-dataset video number only in 364, with the mentioned 992 video does not conform to the paper.
  2. View: The paper mentions that there are four views on the data collected for each patient, but only one view from the publicly available data set.
  3. Annotation : The videos in the "Site_R_73" folder have no annotated data, and the annotated information in the "label_all_frame" folder is polygonal frames instead of regions.
Jiewen-Yang commented 2 months ago

Currently, we provide only view 4 (A4C view) for publicly available, which is why the dataset only contains 364 videos. For the other three views, we are now planning to release them in late of year 2024.

For question 3, the cooperated hospital did not provide the label for Site_R_73. In our experiments, we also report the result under the same situation. We are now communicating with the hospital and asking doctors to annotate the unlabelled data.