How to use the code about "Encoder-Decoder"

[wishinghao]

Hello, I have some questions regarding the code in the "Encoder-Decoder" section. Is it used for training and extracting results? Could you provide a pre-trained model for fitting meshes or images?

[liren2515]

Hello, 'Encoder-Decoder' is used for training and extracting the latent codes (and reconstructed meshes if needed).

The pre-trained models for fitting are already included in the folder checkpoints.

[wishinghao]

When I run the "train-ids.pkl", I get "No such file or directory: 'train-ids.pkl'". How can I get this pkl file?

[wishinghao]

I trained successfully, thank you very much

[wishinghao]

How to train the deformation field? I think the ∆x should be trained, but I can not find reference code.

[liren2515]

The training codes for $\Delta x$ are not provided in this repo. Generally, we follow the physic-based loss from SNUG to train it. The training details can be found in our paper.

[LAW1223]

Thank you for your excellent work, I have some questions about the dataset. The file structure of CLOTH3D is every .obj is belongs to a fold. In this repo, should I put this .obj all together into one fold, according to the preprocess_udf.py

[fatbao55]

@wishinghao I also encountered "No such file or directory: 'train-ids.pkl'". Do you mind sharing how you obtained the train-ids.pkl?

[wishinghao]

I have rewritten the dataloader, so I don't need the pkl file. @fatbao55

[liren2515]

@wishinghao I also encountered "No such file or directory: 'train-ids.pkl'". Do you mind sharing how you obtained the train-ids.pkl?

The training files for Encoder-Decoder are provided here.