Why is the fine-tuning result on modelnet40 obtained by my pre-trained model so different from the pre-trained model you gave?

[ProgrammerSakura]

Here are the results of my fine-tuning:

Model parameter loading process:

The best result I got from reproducing fine-tuning:

You gave:

The model parameter loading process I got using the pre-trained model you gave:

The best result obtained from fine-tuning:

The model parameter loading process is inconsistent. Is it a problem with my pre-training?

The training step script i used is : Pretraining got the pretarin model ckpt-best.pth python main.py --config cfgs/pre-training/point-m2ae.yaml --exp_name pre-train

Then i copy this ckpt-best.pth to ckpts document, and use this scipt to funtune on modelnet40: python main.py --config cfgs/fine-tuning/modelnet40.yaml --finetune_model --exp_name finetune --ckpts ckpts/ckpt-best.pth