Why are inconsistent parameters set in general model config and fine-tune model config?

[EchoTHChen]

(1)
In configs/train/gen/train_gen_cost_volume_train.yaml, use_vis is false.

fine_dist_decoder_cfg:
    use_vis: false

while, in configs/train/ft/neuray_ft_cv_lego.yaml, the use_vis of fine_dist_decoder_cfg is set as the default value of dist_decoder: true.

This will cause the pretrained general model is not successfully loaded in the fine-tune model because of the different setting of vis-encoder in dist-decoder. So what should use_vis of fine_dist_decoder_cfg be ?

(2) I noticed that use_self_hit_prob is only set to true in the fine-tune-model. So why not set it to true consistently in the general-model?

[liuyuan-pal]

1. Hi, both of them are supposed to be set to False. However, the pre-trained model uses the fine_dist_decoder_cfg with use_vis=True. This is a historical problem, in which I used to apply an additional "visibility" (not the visibility of NeuRay) term on the hit probability distribution. Such a "visibility" term almost has no effect on the rendering quality. So I set it to False afterward. However, it seems that I forget to delete them in pretrained models. Sorry for the messy codes.
2. use_self_hit_prob is used in training of fine-tuning not for the generalization model, which means we will decode the hit prob on the pseudo test views during training. In generalization training, such test views are not pseudo but real unseen test views.

[EchoTHChen]

Thanks!