# PixelGen Medical - REFUGE2 Optic Disc/Cup Segmentation
# 3-class mask-conditional fundus image generation
# Dataset: 1200 RGB images (train+val+test), 3-class masks (0/128/255)
# mask_mode=spatial: mask patchified and added to patch embeddings
# Includes null condition dropout for proper CFG
seed_everything: 1234
tags:
  exp: &exp PixelGen_Medical_REFUGE2

trainer:
  default_root_dir: ./medical_workdirs
  accelerator: auto
  strategy: auto
  devices: auto
  num_nodes: 1
  precision: bf16-mixed
  logger:
    class_path: lightning.pytorch.loggers.WandbLogger
    init_args:
      project: pixelgen_medical_refuge2
      name: *exp
  num_sanity_val_steps: 0
  max_steps: 100000
  val_check_interval: 10000
  check_val_every_n_epoch: null
  log_every_n_steps: 50
  deterministic: null
  inference_mode: true
  use_distributed_sampler: false
  callbacks:
    - class_path: src.callbacks.model_checkpoint.CheckpointHook
      init_args:
        every_n_train_steps: 10000
        save_top_k: -1
        save_last: true
    - class_path: src.callbacks.save_images.SaveImagesHook
      init_args:
        save_dir: val_samples
        save_compressed: true
  plugins:
    - src.plugins.bd_env.BDEnvironment

model:
  vae:
    class_path: src.models.autoencoder.pixel.PixelAE
    init_args:
      scale: 1.0
  denoiser:
    class_path: src.models.transformer.JiT_medical.JiTMedical
    init_args:
      input_size: 256
      patch_size: 16
      in_channels: 3
      hidden_size: &hidden_dim 768
      depth: 12
      num_heads: 12
      mlp_ratio: 4.0
      attn_drop: 0.0
      proj_drop: 0.1
      num_classes: 1
      use_bottleneck: true
      bottleneck_dim: 128
      in_context_len: 32
      in_context_start: 4
      mask_in_channels: 1
      mask_mode: spatial
  conditioner:
    class_path: src.models.conditioner.mask_conditioner.MaskConditioner
    init_args:
      hidden_size: *hidden_dim
      in_channels: 1
      img_size: 256
      null_condition_p: 0.1
  diffusion_trainer:
    class_path: src.diffusion.flow_matching.training_medical.MedicalTrainerSimple
    init_args:
      lognorm_t: true
      P_mean: -0.8
      P_std: 0.8
      t_eps: 0.05
      scheduler: &scheduler src.diffusion.flow_matching.scheduling.LinearScheduler
      lpips_weight: 0.1
      percept_t_threshold: 0.3
      null_condition_p: 0.1
  diffusion_sampler:
    class_path: src.diffusion.flow_matching.sampling_medical.EulerSamplerMedical
    init_args:
      num_steps: 50
      guidance: 2.0
      timeshift: 1.0
      guidance_interval_min: 0.1
      guidance_interval_max: 0.9
      scheduler: *scheduler
      w_scheduler: src.diffusion.flow_matching.scheduling.LinearScheduler
      guidance_fn: src.diffusion.base.guidance.simple_guidance_fn
      step_fn: src.diffusion.flow_matching.sampling.ode_step_fn
  ema_tracker:
    class_path: src.callbacks.simple_ema.SimpleEMA
    init_args:
      decay: 0.9999
  optimizer:
    class_path: torch.optim.AdamW
    init_args:
      lr: 1e-4
      weight_decay: 0.0

data:
  train_dataset:
    class_path: src.data.dataset.refuge2.REFUGE2Dataset
    init_args:
      data_root: /data2/sichengli/Data/test/Segmentation/REFUGE2
      resolution: 256
      splits:
        - train
        - val
      augment: true
      val_ratio: 0.1
  eval_dataset:
    class_path: src.data.dataset.refuge2.REFUGE2RandnDataset
    init_args:
      data_root: /data2/sichengli/Data/test/Segmentation/REFUGE2
      resolution: 256
      max_num_instances: 200
  pred_dataset:
    class_path: src.data.dataset.refuge2.REFUGE2RandnDataset
    init_args:
      data_root: /data2/sichengli/Data/test/Segmentation/REFUGE2
      resolution: 256
      max_num_instances: 1000
      noise_scale: 1.0
  train_batch_size: 16
  train_num_workers: 4
  pred_batch_size: 16
  pred_num_workers: 1