train

[wudami]

Hello, I have a couple of questions I would like to ask. ① Can you tell me if the training model is to run the train_hcell.py file first, and then run the train_tcell.py file? ② Is it necessary to add the path to the configuration file (yaml) in the code 'parser.add_argument('--config')'? ③ Is it possible to provide the detailed steps for training and testing?

[minshu-kim]

1. Yes but Not absolutely. You can skip the training of H-Cell with train_hcell.py if you use a pre-trained H-Cell model.
2. Yes.
3. REwarp consists of a Homography estimator (H-Cell) and a TPS estimator (T-Cell) for associating elastic warps. "train_hcell.py" corresponds to the training of the homography estimator H-Cell, and "train_tcell.py" contains T-Cell's training. Our test steps are in "eval.py". The codes contain REwarp's estimation of elastic warps by associating grid deformations (or warp fields) with the estimated homography and TPS. You can see more details in our paper.

[wudami]

1. 是的，但不是绝对的。如果您使用预先训练的 H-Cell 模型，则可以跳过带有 train_hcell.py 的 H-Cell 训练。
2. 是的。
3. REwarp 由用于关联弹性翘曲的单调估计器 （H-Cell） 和 TPS 估计器 （T-Cell） 组成。“train_hcell.py”对应单调估计器H-Cell的训练，“train_tcell.py”包含T-Cell的训练。我们的测试步骤是“eval.py”。REwarp 通过将网格变形（或翘曲场）与估计的单调和 TPS 相关联来估计弹性翘曲。您可以在我们的论文中看到更多详细信息。

Thanks for getting back to me, also I have another question, what is the eval_psnr.py file used for?

[minshu-kim]

eval_psnr.py codes measure masked PSNR (mPSNR) of aligned images. eval.py codes additionally contain the blending and stitching of the aligned images.

[wudami]

eval_psnr.py codes measure masked PSNR (mPSNR) of aligned images. eval.py codes additionally contain the blending and stitching of the aligned images.

Thank you very much for your answer.