The phenomenon of confusion in predicting targets during inference occurs after continuing training on a self built dataset

[Unicorn123455678]

Will the parameters of the CLIP text editor be updated when fine-tuning the self built COCO dataset? When I was using Simple_demo.exe in the demo folder and used/YOLO World master/tools/work-dirs/xxx/epoch-80. pth (fine-tuning model parameters from my self built dataset) as a checkpoint, I found that it still seemed to be the parameters loaded from the CLIP native text editor during debugging. This may be the fundamental reason for my model prediction confusion. I hope the author can provide an answer!

[lvke9529]

Have you solved this problem?