EllipssianNet

0. Pre-Requisite

conda env create -f environment.yml
conda activate ellipssianNet

1. Run (Inference)

python train.py --img_path "C:/your_img_path" 

or

bash run.sh 

2. Create Dataset

python create_dataset.py --save_path "C:/your_dataset_path"

or

bash create_dataset.sh

Voronoi diagram (used as GT)	Edges of the diagram

Gradient map (used as GT)	Center probability map (used as GT)	Rendered Ellipssians

2.1 Basic parameters

• --batch
  • Number of dataset to be created (100 by default)
  • python create_dataset.py --batch 200
• --render
  • Visualize(or not) the created dataset (True by default)
  • python create_dataset.py --render True
  • python create_dataset.py --render False
• --save_path
  • Path to store the created dataset (Blank by default - does not save)
  • python create_dataset.py --save_path "C:/your_path"

2.2 Batch command parameters

• --iteration
  • Considers the code is running with nth iteration (0 by default)
  • Indices of dataset are computed with this parameter.
• --begin_batch
  • Beginning index number (0 by default)

2.3 Usage in .sh file

• The following command runs the code 10 (0-9) iterations. In total, dataset size of 2000 is to be created.

  for (( x=0; x<=9; x++ ))
  do
     iter=$((x+1))
     echo "Running iteration $iter/3"
     python create_dataset.py --iteration $x --batch 200 
  done
  

• The following command creates dataset size of 190 (200 - 10), beginning with 10, ending with 199.

     python create_dataset.py --batch 200 --begin_batch 10
  

---

2.4 For whom may be curious

• The following command creates dataset size of 1900 ((200 - 10) * 10).
• The indices of dataset will be 10-199, 210-399, 410-599, etc.

  for (( x=0; x<=9; x++ ))
  do
     iter=$((x+1))
     echo "Running iteration $iter/3"
     python create_dataset.py --iteration $x --batch 200 --begin_batch 10 
  done
  

---

3. Train

python train.py  --dataset_path "C:/your_path" --chkpoint_save_path "C:/your_path" --epoch 100 --dataset_num 50000 --batch_size 10