Once the data is preprocessed, we can now train the model and infer from it. Instructions are given below:
python train.py --data <data_dir> --checkpoint <ckpt_dir> --logdir <log_dir> --epoch 100 \
--load <lif_ckpt_load_path> --load_step <lif_ckpt_epoch> --mask_load <mask_ckpt_load_path> --mask_load_step <mask_ckpt_epoch>
You can tweak other parameters like batch_size, etc in config.py
- For real scenes
python render_network.py --data <data_dir> --lif_checkpoint <lif_ckpt_path> --mask_checkpoint <mask_ckpt_path> --output_dir <output_path> --material <input_material>
This implementation is an unofficial version of Zp Zhou. The code can be found here