High-Efficiency Classification of White Blood Cells Based on Object Detection

Figure 6

Training output of Yolov4 and Faster RCNN. After 100 epochs, Faster RCNN’s loss value curve (a) drops to 0.2, while Yolov4’s loss value (b) drops to 0.02. Yolov4’s precision and recall rate curve fluctuate widely, and Faster RCNN’s precision curve is smoother, but the recall rate is about 10 percentage points lower than Yolov4’s (c, d).

(a)

(b)

(c)

(d)