The confidence level of multiple class models is lower than that of a single model

[Nanmei]

I use weigth of yolov4.conv.137 to training
max_batches = 5000
I trained a class of models with 1700 images ， one class detection effect, with a confidence level of 0.98

voc-dp.data.txt
yolov4-custom.cfg.txt
yolov4-custom-test.cfg.txt

When I added 2 classes, these two classes are associated with 1400 images

gap is where I added 2 classes of corresponding images,I added 2 classes of 1400 images.
Now I have 3 classes.I use weigth of yolov4.conv.137 to training
max_batches = 50000

voc-dp.data.txt

yolov4-custom.cfg.txt

yolov4-custom-test.cfg.txt

I use the weights obtained from training three classes to detect. The confidence level of a class is very low, with a result of 0.44. Even other images cannot be detected

The training of multiple classes has caused my detection confidence to be very low. I cannot find the reason and cannot solve this problem at present