mtonym / Dodekathlon

pytorch-1.3.0 training-dev

Dodekathlon

Updates:

• 2019-11-28 Hydra version under construction.
• 2019-11-12 Name of my projects will be changed to the Myths from Ancient Greek.

Hydra Checklist.

• Checkpoints support.
• kp dataset/ kpparse dataset
• model for PATN
• training
• losses
• evaluations.

Canary Plans.

• Example of PATN and Controllable Human Pose Transfer branch:hydra
• Distributed training process.
• Visualization with TensorBoard embedded.
• Deploying using oonx and TensorRT

Citation

if you find this code benefits your research, please take the citation at ease:

@inproceedings{zhu2019progressive,
  title={Progressive Pose Attention Transfer for Person Image Generation},
  author={Zhu, Zhen and Huang, Tengteng and Shi, Baoguang and Yu, Miao and Wang, Bofei and Bai, Xiang},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  pages={2347--2356},
  year={2019}
}

Structure of Dodekathlon

├── configures
│   ├── train.yaml // The configures. (Trail Level)
│   └── ....    // store the yaml configure.
├── criterions  // define the criterions train and test.
│   ├── loss    // define the loss.
│   └── metrics // define the metrics for evaluation.
├── datasets    // define the dataset. dataloader.
├── deployments // deploy to the cloud using oonx.
├── main.py     // great dodekathlon main file.
├── models      // define the model here.
├── options     // define the options. DIFFS From configures.(Task level)
├── requirements.txt // try 'pip install -Ur requirements.txt'
├── tools       // some useful tools(object oriented).
└── utils       // some useful functions.

Reference of Hercules Dodekathlon

Before I chose this name "Dodekathlon", I supposed to use Hercules, who was considered as a demi-god with strong labors. "Dodekathlon", considered as 12 Labors of Hercules, are how Hercules gained much of his mythological fame as a demi-god, listing as follows:

(picture credit to wikipedia)

• Slay the Nemean Lion.
• Slay the nine-headed Lernaean Hydra.
• Capture the Golden Hind of Artemis.
• Capture the Erymanthian Boar.
• Clean the Augean stables in a single day.
• Slay the Stymphalian Birds.
• Capture the Cretan Bull.
• Steal the Mares of Diomedes.
• Obtain the girdle of Hippolyta, Queen of the Amazons.
• Obtain the cattle of the monster Geryon.
• Steal the apples of the Hesperides
• Capture and bring back Cerberus.

As PyTorch is from Torch, whose meaning is a sticker with flames and fires. This project combines Training, Testing, Deploying, Visualization, Evaluation, Checkpoints and Resuming, etc., each of them is challenging and in great modularization. Therefore, I chose Dodekathlon.