Lightweight overview of the full pipeline: prepare paired H&E/IHC WSIs, review them in a Dash viewer, and train a Pix2Pix model for stain translation.

• Full walkthrough is in presi/presi.pdf (project context + visuals).

• data_prep/: WSIWorker pipeline for segmentation, manual selection, and VALIS registration (+ helper scripts in run_code/).
• dash-plotly-wsi-viewer/: Dash app to browse tiled WSIs, pair IHC↔H&E segments, and collect alignment ratings.
• src/: Training and inference (PyTorch Lightning + Hydra) for Pix2Pix on paired patches.
• conf/: Hydra configs for paths, model, training, augmentations, and WandB.
• presi/: Slides for the project (PowerPoint + PDF) once uploaded.

• Copy conf/wandb/wandb.yaml.example to conf/wandb/wandb.yaml.
• Fill in your WandB entity/project (or disable via wandb.enabled=false in Hydra).

• Core pipeline (no Dash server): use the root-level requirements.txt.
  • Submodules data_prep/trident_sinclair, data_prep/mapping_frameworks/valis, and (optional) data_prep/mapping_frameworks/DeeperHistReg are not bundled in requirements.txt; install them yourself from the checked-out submodules, e.g.:
    • git submodule update --init data_prep/trident_sinclair data_prep/mapping_frameworks/valis
    • pip install -e data_prep/trident_sinclair
    • pip install -e data_prep/mapping_frameworks/valis
    • (optional) git submodule update --init data_prep/mapping_frameworks/DeeperHistReg && pip install -e data_prep/mapping_frameworks/DeeperHistReg
• Dash server: will ship with a Dockerfile in dash-plotly-wsi-viewer/ to build/run the viewer in isolation; use that instead of the core dash-plotly-wsi-viewer/requirements.txt.

1. Prep WSIs in data_prep/
   • Segment tissue (WSIWorker.segment_*), pick relevant regions, and register IHC→H&E with VALIS via the batch scripts in data_prep/run_code/ (SimpleElastix + OpticalFlow outputs).
2. Serve & review in dash-plotly-wsi-viewer/
   • Tile registered TIFFs, create on_server.csv/pairs.json, and run the Dash server to rate alignments and mark approved pairs.
3. Train Pix2Pix in src/
   • Point Hydra paths to the processed dataset, then run python src/train.py (optionally wandb.enabled=false). Use src/inference.py for checkpoints.

See the dedicated READMEs in each folder for details.