docsom/topdig

TopDiG (CVPR 2023) paper implementation

Relationformer: A Unified Framework for Image-to-Graph Generation

Requirements

• CUDA>=9.2
• PyTorch>=1.7.1

For other system requirements please follow

pip install -r requirements.txt

Compiling CUDA operators

cd ./models/ops
python setup.py install

Code Usage

1. Dataset preparation

Please download 20 US Cities dataset and organize them as following:

code_root/
└── data/
    └── 20cities/

After downloading the dataset run the following script to preprocess and prepare the data for training

python generate_data.py

2. Training

2.1 Prepare config file

The config file can be found at .configs/road_rgb_2D.yaml. Make custom changes if necessary.

2.2.a Training on multiple-GPU (e.g. 3 GPUs)

For example, the command for training Relationformer is following:

python train.py --config configs/road_rgb_2D.yaml --cuda_visible_device 0 1 2 --nproc_per_node 3

3. Evaluation

Once you have the config file and trained model of Relation, run following command to evaluate it on test set:

python test.py --config configs/road_rgb_2D.yaml --checkpoint ./trained_weights/last_checkpoint.pt

4. Interactive notebook

Please find the debug_relationformer.ipynb for interactive evaluation and visualization