vanillacoke/meddreambooth-structure

MedDreamBooth Journal Paper

Repository containing the complete MedDreamBooth project for journal submission.

Structure

• meddreambooth_journal/: Main project directory

  • data/: Training data (images, masks, splits)
  • scripts/: Training and evaluation scripts
  • multiseed_results/: Trained models (exp1 and exp22)
  • multiseed_evaluation/: Evaluation results (Table 3)
  • held_out_evaluation/: Held-out evaluation results (Table 5)
  • pretrained_models/: Pre-trained segmentation model

• train_1.py: Main training script (fixed version)

• train_multiseed.py: Multi-seed training wrapper

Quick Start

# Setup
pip install -r requirements.txt
huggingface-cli login

# Training
python meddreambooth_journal/scripts/train_multiseed.py \
  --instance_data_dir meddreambooth_journal/data/instance_images/images \
  --instance_mask_dir meddreambooth_journal/data/instance_masks \
  --class_data_dir meddreambooth_journal/data/class_images \
  --seg_model_weights meddreambooth_journal/pretrained_models/best_segmentation_model_50_epochs.pth \
  --output_base_dir meddreambooth_journal/multiseed_results \
  --num_seeds 5 --seeds 0 42 123 999 12345 \
  --exp1_epochs 10 --exp22_epochs 40 --gpu_id 0

# Evaluation
python meddreambooth_journal/scripts/evaluate_multiseed.py \
  --results_root meddreambooth_journal/multiseed_results \
  --real_images_path meddreambooth_journal/data/eval_images_unseen \
  --num_samples 1000 --batch_size 4 \
  --output_dir meddreambooth_journal/multiseed_evaluation \
  --prompt "a photo of <nih-xray>"

Results

• Table 3: meddreambooth_journal/multiseed_evaluation/multiseed_summary.csv
• Table 5: meddreambooth_journal/held_out_evaluation/

Key Files

• train_1_fixed.py: Training script with all fixes applied
• train_multiseed.py: Multi-seed wrapper with auto-resume
• CHAT_SUMMARY_FOR_NEXT_CLAUDE.md: Setup and progress summary

Requirements

See requirements.txt for dependencies.

---

For more details, see CHAT_SUMMARY_FOR_NEXT_CLAUDE.md