Different Results When testing On NeuS.

[MobiusLqm]

Hi, benny! I meet a strange error. I tried to train NeuS on several scenes. After training, your codes will automatical excute the evalueation like the codes below. if args.train: if args.resume and not args.resume_weights_only: # FIXME: different behavior in pytorch-lighting>1.9 ? trainer.fit(system, datamodule=dm, ckpt_path=args.resume) else: trainer.fit(system, datamodule=dm) trainer.test(system, datamodule=dm) However, When I trying to do evaluation on by loading the saved ckpt after training and set --test in cmd argument, the metrics are quiet different from the results reported after training,for example, PSNR is only 32.00 compared with 35.00 tested after training immediately. Do you have any idea about this?