During an attempt to duplicate a project in the RoboFlow system, we identified a critical bug that affected one of the datasets in the “Annoting” section of the base project (not the duplicated one). In the base project, one of the datasets in the “Annoting” section had a total of 18 images, with 7 marked as “annotated” and 11 as “unannotated.” However, after duplicating the project and making changes ONLY in the copy of that project, out of the 7 “annotated” images, 2 were moved to “unannotated,” and the rest were transferred to the dataset of the copied project. However, upon returning to the base project, we noticed that the section, which should originally have had 7 images marked as “annotated” and 11 as “unannotated,” now erroneously reports that this distribution is correct. However, upon accessing the section, we found only 5 images marked as “annotated” and no images marked as “unannotated.”
- Action Taken: Attempt to duplicate an existing project in the RoboFlow platform.
- Affected Section: “Annoting” of the base project.
- Bug Description: After the successful duplication of the project, the distribution of images in the “Annoting” section of the base project did not accurately reflect the original information. The system erroneously reported that the images were distributed as they should be, but, in reality, only 5 images were marked as “annotated,” and none were marked as “unannotated.” Furthermore, in the copied project, there was an unexpected change in the annotation category of 2 images.
- Impact: The bug resulted in an incorrect distribution of images between the “annotated” and “unannotated” categories in the base project, compromising data integrity and project continuity.
Question Regarding Data Recovery: We would like to inquire whether the RoboFlow team has the capability to restore the correct distribution of images in the “annotated” and “unannotated” categories in the base project. Additionally, it would be helpful to understand how to prevent this type of bug from occurring in the future to maintain dataset accuracy.
base project: Sign in to Roboflow