Is this updated dataset_1 any different from the previous ones? And is it necessary to update it?
Created by zou baosheng zoubaosheng The videos and action recognition annotations are the same.
The updated training_set_1 includes the following correction to the segmentation labels:
- the addition of frame 000040020.png in video_6 which was missing from the first version
- corrections in the thread and clamp labels
- corrections in the tool masks, where, initially, blood blobs accumulated underneath tools defined the contour of the tools' mask. In the new version, the shape of tool masks is roughly estimated based on the tool shape. Datasets 4, 16 and 19 bring such changes.
We recommend updating the training_set_1 because it has fewer errors and retraining with the new version may help increase the accuracy of your models.