RG-DermNet

This repository is focused on the experimental pipeline used in the scientific paper:

• multimodal training scripts in src/scripts/benchmark
• metric aggregation and statistical analysis in src/scripts/aggreation

1. Environment setup

conda create -n rg-dermnet python=3.10 -y
conda activate rg-dermnet
pip install -r requirements.txt

2. Configuration (conf/.env)

Training scripts read the following variables from conf/.env:

• NUM_EPOCHS
• BATCH_SIZE
• K_FOLDS
• LIST_NUM_HEADS
• COMMON_DIM
• DATASET_FOLDER_NAME
• DATASET_FOLDER_PATH
• RESULTS_FOLDER_PATH
• NUMBER_OF_WORKERS
• UNFREEZE_WEIGHTS
• LLM_MODEL_NAME_SEQUENCE_GENERATOR
• save_to_disk

Expected dataset structure:

<DATASET_FOLDER_PATH>/
|-- images/
`-- metadata.csv

3. Training

Example runs:

python src/scripts/benchmark/train_pad_20.py
python src/scripts/benchmark/train_pad_25.py
python src/scripts/benchmark/train_isic_2019.py
python src/scripts/benchmark/train_isic_2020.py

Metrics per fold are saved under RESULTS_FOLDER_PATH.

4. Aggregation and statistics

After training:

python src/scripts/aggreation/average_metric_values.py
python src/scripts/aggreation/statistical_test.py

Main files:

• src/scripts/aggreation/average_metric_values.py
• src/scripts/aggreation/statistical_test.py
• src/scripts/aggreation/stats.py

5. Minimum project scope

For paper reproducibility, the core is:

• conf/
• requirements.txt
• src/scripts/benchmark/
• src/scripts/aggreation/
• data/ (local, not versioned)

Citation (IJCNN 2026)

If you use this project, please cite:

Rocha, W. F., Bouzon, P. H. G., Ramos, L. A., Pacheco, A. G. C., and Souza Jr., L. A.
RG-DermNet: A Multimodal Attention-Based Model with Residual Block Usage for Skin Lesion Classification.
Accepted at the International Joint Conference on Neural Networks (IJCNN), 2026.

Authors and affiliations

• Wyctor F. da Rocha, Pedro H. G. Bouzon, Andre G. C. Pacheco, Luis A. Souza Jr.
  Graduate Program of Informatics, Federal University of Espirito Santo, Vitoria, Brazil
• Lucas A. Ramos
  Computer Vision and Data Science, NHL Stenden University of Applied Sciences, Leeuwarden, The Netherlands

BibTeX

@inproceedings{rocha2026rgdermnet,
  title     = {RG-DermNet: A Multimodal Attention-Based Model with Residual Block Usage for Skin Lesion Classification},
  author    = {Rocha, Wyctor F. and Bouzon, Pedro H. G. and Ramos, Lucas A. and Pacheco, Andre G. C. and Souza Jr., Luis A.},
  booktitle = {International Joint Conference on Neural Networks (IJCNN)},
  year      = {2026},
  note      = {Accepted}
}

Questions and support

For further questions about the lab and related projects, please visit:

https://life.inf.ufes.br/