This is a relabeled BCCD dataset in Pascal VOC format.
In view of the large number of error and missing labels in the original BCCD dataset, I spent the whole day re-labeling each picture. Using this data set, the accuracy of retraining on the yolov5x pre-train model is greatly improved.