About unsupervised losses

[hyalvin]

Hi, in your provided training details in hugging face, I saw that all the training logs show zero unsupervised losses in the beginning several iterations. Why this would happen?

At the same time, when I apply MixPL to YoloV8 on my custom dataset, if I initialize the detector'weights using the supervised-trained baseline, the training process seems to be unstable and the model seems to be degraded quickly. Which, then, leads to an eval error like below

loading annotations into memory...
Done (t=0.59s)
creating index...
index created!
Loading and preparing results...
Traceback (most recent call last):
  ***
  File "/***/mmyolo/lib/python3.8/site-packages/pycocotools/coco.py", line 329, in loadRes
    if 'caption' in anns[0]:
IndexError: list index out of range

Seems to be there is no prediction on validation dataset at all. I would appreciate it if you can take a look at this problem, thanks!

[Czm369]

By default, each unlabeled image contains at least one object. In the early stage of training, the detection ability of the model is weak, and the confidence of all the predicted results is lower than the threshold, so the loss is not calculated