Reproduction codes of Causal Info GAN with chainer
See this original paperl for details
Download rope_full and seq_data_2 datasets from author's github repo and place the datasets under datasets directory as below
datasets
├── rope_full
│ ├── run05
│ ├── run06
│ ...
└── seq_data_2
├── goal
│ └── all
│ ├── img_0000_g.jpg
│ ...
└── start
└── all
├── img_0000_s.jpg
...
| rope state | next rope state |
|---|---|
 |
 |
Work is in progress... Rope images are successfully generated but planning still fails. See below
These images can be obtained with below command
$ python3 tester.py --generator-parameter=trained_models/generator_epoch-100 --posterior-parameter=trained_models/posterior_epoch-100 --subtractor-parameter=trained_models/subtractor.model --transition-parameter=trained_models/transision_epoch-100 --classifier-parameter=trained_models/classifier.model --start-image=datasets/seq_data_2/start/all/img_0000_s.jpg --goal-image=datasets/seq_data_2/goal/all/img_0000_g.jpg