Code adapted from Aladdin Persson.
This is my project on Deep Learning. It generates traditional thai house images using DCGANs. Total training images = 1,000 images (10,000 after augmented). The result images are still not good. More training, data, and tuning needed.
├───dataset
│ └───[your_dataset]
└───images
├───gen_img
│ └───[name_your_folder]
├───saved_models
│ └───[name_your_folder]
└───tensorboard_logs
└───[name_your_folder]
See section 1, 2 in the notebook
- Put training images in
dataset/[your_dataset]/images/
- Set project path in the notebook
DATASET_PATH = "./dataset/[your_dataset]" PROJECT_PATH = "[name_your_folder]"
- Run
Thaihouse-gen_DCGANs.ipynb
- Weights will be saved every epochs in
saved_models/
- Tensorboard logs will be in
tensorboard_logs/
Training with 50epochs
Tensorboard
See section 3, 4 in the notebook. Generate one random tensor and put into generator.
See section 5 in the notebook. But as you can see, generator loss is still not converging.