Torch implementation of SDF-GCQN.
Figure 1. The overview of SDF-GCQN.
SDF-GCQN consists of two parts: a scene graph generator and a scene graph encoder. The scene graph generation consists of merging two scene subgraphs. The subgraph of each scene is a complete graph using the SDFs of objects as nodes. The scene graph encoder consists of CNN layer-based graph convolution layers.
To train a model:
# Rendering on #
python dqn_train.py --render --show_q
# Rendering off #
python dqn_train.py --gpu [GPU_ID] --show_q
To test the trained model:
# Rendering on #
python dqn_eval.py --model_path [MODEL_NAME] --render --show_q
# Rendering off #
python dqn_eval.py --model_path [MODEL_NAME] --gpu [GPU_ID] --show_q
To test the rule-based method:
# Rendering on #
python rulebased_eval.py --render --show_q
# Rendering off #
python rulebased_eval.py --gpu [GPU_ID] --show_q
MIT License.