tigerqiu712 / synz

Enhanced Synthetic Dataset for Training UI Element Detector From Lo-Fi Sketches

SynZ: Enhanced Synthetic Dataset for Training UI Element Detector From Lo-Fi Sketches

💾 SynZ Dataset in Kaggle      💾 UISketch Dataset in Kaggle

---

Script to extract SynZ annotations from RICO

• Follow the steps and run the script from the notebooks folder
  • Notebook RICO_data_extraction.ipynb converts RICO annotations to a CSV file
  • Notebook RICO_to_SynZ_annotations.ipynb converts RICO annotations to SynZ annotations for further steps

---

Generating SynZ sketches

To generate SynZ sketches from scratch,

• Download all the UISketch dataset to the data folder

  • UISketch dataset from kaggle

• Either run the notebooks described in previous section to generate SynZ_ready_annotations.csv file or download the SynZ annotations from here

• Install the necessary requirements from requirements.txt file

  pip install -r requirements.txt

• Run the following command to generate SynZ dataset

  python generate_synz_dataset.py \
    --annotation_csv_path=./data/SynZ_ready_annotations.csv \
    --uisketch_labels_csv_path=./data/uisketch/labels.csv \
    --output_path=./data/SynZ \
    --number_of_versions=1 \
    run
  

---

Downlading MetaMorph model trained weights

• Download the trained SynZ checkpoint from here to models folder
• Extract the synz_checkpoint.tar.gz folder to acquire the MetaMorph model's trained weights from model_final.pth

---

Authors

👤 Vinoth Pandian

• Website: vinoth.info
• Github: @vinothpandian
• LinkedIn: @vinothpandian

👤 Sarah Suleri

• Website: sarahsuleri.info
• LinkedIn: @sarahsuleri