feat(dynacell): segmenter-consistency fine-tune loss (P1, nucleus)

applications/dynacell/configs/benchmarks/virtual_staining/nucleus/fcmae_vscyto3d_scratch_segcons/ipsc_confocal/predict__a549_mantis_denv.yml

@@ -0,0 +1,44 @@
+# GENERATED by applications/dynacell/tools/generate_spotlight_v2_leaves.py -- edit the
+# generator, not this file (plan: ~/.claude/plans/spotlight-v2.md).
+# Spotlight-v2 `segcons` arm (predict) of the baseline leaf nucleus/fcmae_vscyto3d_scratch/ipsc_confocal/predict__a549_mantis_denv.yml.
+# Inference is identical to the baseline's; only the checkpoint, store and names differ.
+# Fine-tune arm: predict the end of the fine-tune; submit with `--ckpt last`.
+# Everything else, and the rationale for it, is the baseline leaf's.
+base:
+- ../../../_internal/shared/model/predict_sets/a549_mantis_h2b_denv.yml
+- ../../../_internal/shared/model/targets/nucleus.yml
+- ../../../_internal/shared/model/model_overlays/fcmae_vscyto3d_predict.yml
+- ../../../_internal/shared/model/launcher_profiles/mode_predict.yml
+- ../../../_internal/shared/model/launcher_profiles/hardware_predict_any_gpu.yml
+- ../../../_internal/shared/model/launcher_profiles/runtime_shared.yml
+benchmark:
+  task: virtual_staining
+  organelle: nucleus
+  trained_on: ipsc_confocal
+  predict_set: a549_mantis_h2b_denv
+  model_name: fcmae_vscyto3d_scratch_segcons
+  experiment_id: nucleus__ipsc_confocal__fcmae_vscyto3d_scratch_segcons__a549_mantis_h2b_denv
+  dataset_ref:
+    target: h2b
+model:
+  init_args:
+    ckpt_path: /hpc/projects/comp.micro/virtual_staining/models/dynacell/ipsc/nucleus/fcmae_vscyto3d_scratch_segcons/checkpoints/last.ckpt
+data:
+  init_args:
+    normalizations:
+    - class_path: viscy_transforms.NormalizeSampled
+      init_args:
+        keys:
+        - Phase3D
+        level: fov_statistics
+        subtrahend: mean
+        divisor: std
+    augmentations: []
+trainer:
+  callbacks:
+  - class_path: viscy_utils.callbacks.prediction_writer.HCSPredictionWriter
+    init_args:
+      output_store: /hpc/projects/virtual_staining/training/dynacell/nucleus/fcmae_vscyto3d_scratch_segcons/ipsc/a549__denv/prediction.zarr
+launcher:
+  job_name: FCMAE_VSCyto3D_Scratch_PRED_NUCL_ON_A549_DENV_segcons
+  run_root: /hpc/projects/virtual_staining/training/dynacell/nucleus/fcmae_vscyto3d_scratch_segcons/ipsc/a549__denv

applications/dynacell/configs/benchmarks/virtual_staining/nucleus/fcmae_vscyto3d_scratch_segcons/ipsc_confocal/predict__a549_mantis_mock.yml

@@ -0,0 +1,44 @@
+# GENERATED by applications/dynacell/tools/generate_spotlight_v2_leaves.py -- edit the
+# generator, not this file (plan: ~/.claude/plans/spotlight-v2.md).
+# Spotlight-v2 `segcons` arm (predict) of the baseline leaf nucleus/fcmae_vscyto3d_scratch/ipsc_confocal/predict__a549_mantis_mock.yml.
+# Inference is identical to the baseline's; only the checkpoint, store and names differ.
+# Fine-tune arm: predict the end of the fine-tune; submit with `--ckpt last`.
+# Everything else, and the rationale for it, is the baseline leaf's.
+base:
+- ../../../_internal/shared/model/predict_sets/a549_mantis_h2b_mock.yml
+- ../../../_internal/shared/model/targets/nucleus.yml
+- ../../../_internal/shared/model/model_overlays/fcmae_vscyto3d_predict.yml
+- ../../../_internal/shared/model/launcher_profiles/mode_predict.yml
+- ../../../_internal/shared/model/launcher_profiles/hardware_predict_any_gpu.yml
+- ../../../_internal/shared/model/launcher_profiles/runtime_shared.yml
+benchmark:
+  task: virtual_staining
+  organelle: nucleus
+  trained_on: ipsc_confocal
+  predict_set: a549_mantis_h2b_mock
+  model_name: fcmae_vscyto3d_scratch_segcons
+  experiment_id: nucleus__ipsc_confocal__fcmae_vscyto3d_scratch_segcons__a549_mantis_h2b_mock
+  dataset_ref:
+    target: h2b
+model:
+  init_args:
+    ckpt_path: /hpc/projects/comp.micro/virtual_staining/models/dynacell/ipsc/nucleus/fcmae_vscyto3d_scratch_segcons/checkpoints/last.ckpt
+data:
+  init_args:
+    normalizations:
+    - class_path: viscy_transforms.NormalizeSampled
+      init_args:
+        keys:
+        - Phase3D
+        level: fov_statistics
+        subtrahend: mean
+        divisor: std
+    augmentations: []
+trainer:
+  callbacks:
+  - class_path: viscy_utils.callbacks.prediction_writer.HCSPredictionWriter
+    init_args:
+      output_store: /hpc/projects/virtual_staining/training/dynacell/nucleus/fcmae_vscyto3d_scratch_segcons/ipsc/a549__mock/prediction.zarr
+launcher:
+  job_name: FCMAE_VSCyto3D_Scratch_PRED_NUCL_ON_A549_MOCK_segcons
+  run_root: /hpc/projects/virtual_staining/training/dynacell/nucleus/fcmae_vscyto3d_scratch_segcons/ipsc/a549__mock

applications/dynacell/configs/benchmarks/virtual_staining/nucleus/fcmae_vscyto3d_scratch_segcons/ipsc_confocal/predict__a549_mantis_zikv.yml

@@ -0,0 +1,44 @@
+# GENERATED by applications/dynacell/tools/generate_spotlight_v2_leaves.py -- edit the
+# generator, not this file (plan: ~/.claude/plans/spotlight-v2.md).
+# Spotlight-v2 `segcons` arm (predict) of the baseline leaf nucleus/fcmae_vscyto3d_scratch/ipsc_confocal/predict__a549_mantis_zikv.yml.
+# Inference is identical to the baseline's; only the checkpoint, store and names differ.
+# Fine-tune arm: predict the end of the fine-tune; submit with `--ckpt last`.
+# Everything else, and the rationale for it, is the baseline leaf's.
+base:
+- ../../../_internal/shared/model/predict_sets/a549_mantis_h2b_zikv.yml
+- ../../../_internal/shared/model/targets/nucleus.yml
+- ../../../_internal/shared/model/model_overlays/fcmae_vscyto3d_predict.yml
+- ../../../_internal/shared/model/launcher_profiles/mode_predict.yml
+- ../../../_internal/shared/model/launcher_profiles/hardware_predict_any_gpu.yml
+- ../../../_internal/shared/model/launcher_profiles/runtime_shared.yml
+benchmark:
+  task: virtual_staining
+  organelle: nucleus
+  trained_on: ipsc_confocal
+  predict_set: a549_mantis_h2b_zikv
+  model_name: fcmae_vscyto3d_scratch_segcons
+  experiment_id: nucleus__ipsc_confocal__fcmae_vscyto3d_scratch_segcons__a549_mantis_h2b_zikv
+  dataset_ref:
+    target: h2b
+model:
+  init_args:
+    ckpt_path: /hpc/projects/comp.micro/virtual_staining/models/dynacell/ipsc/nucleus/fcmae_vscyto3d_scratch_segcons/checkpoints/last.ckpt
+data:
+  init_args:
+    normalizations:
+    - class_path: viscy_transforms.NormalizeSampled
+      init_args:
+        keys:
+        - Phase3D
+        level: fov_statistics
+        subtrahend: mean
+        divisor: std
+    augmentations: []
+trainer:
+  callbacks:
+  - class_path: viscy_utils.callbacks.prediction_writer.HCSPredictionWriter
+    init_args:
+      output_store: /hpc/projects/virtual_staining/training/dynacell/nucleus/fcmae_vscyto3d_scratch_segcons/ipsc/a549__zikv/prediction.zarr
+launcher:
+  job_name: FCMAE_VSCyto3D_Scratch_PRED_NUCL_ON_A549_ZIKV_segcons
+  run_root: /hpc/projects/virtual_staining/training/dynacell/nucleus/fcmae_vscyto3d_scratch_segcons/ipsc/a549__zikv

applications/dynacell/configs/benchmarks/virtual_staining/nucleus/fcmae_vscyto3d_scratch_segcons/ipsc_confocal/predict__ipsc_confocal.yml

@@ -0,0 +1,42 @@
+# GENERATED by applications/dynacell/tools/generate_spotlight_v2_leaves.py -- edit the
+# generator, not this file (plan: ~/.claude/plans/spotlight-v2.md).
+# Spotlight-v2 `segcons` arm (predict) of the baseline leaf nucleus/fcmae_vscyto3d_scratch/ipsc_confocal/predict__ipsc_confocal.yml.
+# Inference is identical to the baseline's; only the checkpoint, store and names differ.
+# Fine-tune arm: predict the end of the fine-tune; submit with `--ckpt last`.
+# Everything else, and the rationale for it, is the baseline leaf's.
+base:
+- ../../../_internal/shared/model/predict_sets/ipsc_confocal.yml
+- ../../../_internal/shared/model/targets/nucleus.yml
+- ../../../_internal/shared/model/model_overlays/fcmae_vscyto3d_predict.yml
+- ../../../_internal/shared/model/launcher_profiles/mode_predict.yml
+- ../../../_internal/shared/model/launcher_profiles/hardware_predict_any_gpu.yml
+- ../../../_internal/shared/model/launcher_profiles/runtime_shared.yml
+benchmark:
+  task: virtual_staining
+  organelle: nucleus
+  trained_on: ipsc_confocal
+  predict_set: ipsc_confocal
+  model_name: fcmae_vscyto3d_scratch_segcons
+  experiment_id: nucleus__ipsc_confocal__fcmae_vscyto3d_scratch_segcons__ipsc_confocal
+model:
+  init_args:
+    ckpt_path: /hpc/projects/comp.micro/virtual_staining/models/dynacell/ipsc/nucleus/fcmae_vscyto3d_scratch_segcons/checkpoints/last.ckpt
+data:
+  init_args:
+    normalizations:
+    - class_path: viscy_transforms.NormalizeSampled
+      init_args:
+        keys:
+        - Phase3D
+        level: fov_statistics
+        subtrahend: mean
+        divisor: std
+    augmentations: []
+trainer:
+  callbacks:
+  - class_path: viscy_utils.callbacks.prediction_writer.HCSPredictionWriter
+    init_args:
+      output_store: /hpc/projects/virtual_staining/training/dynacell/nucleus/fcmae_vscyto3d_scratch_segcons/ipsc/ipsc/prediction.zarr
+launcher:
+  job_name: FCMAE_VSCyto3D_Scratch_PRED_NUCL_segcons
+  run_root: /hpc/projects/virtual_staining/training/dynacell/nucleus/fcmae_vscyto3d_scratch_segcons/ipsc/ipsc

applications/dynacell/configs/benchmarks/virtual_staining/nucleus/fcmae_vscyto3d_scratch_segcons/ipsc_confocal/train.yml

@@ -0,0 +1,71 @@
+# GENERATED by applications/dynacell/tools/generate_spotlight_v2_leaves.py -- edit the
+# generator, not this file (plan: ~/.claude/plans/spotlight-v2.md).
+# Spotlight-v2 `segcons` arm (fit) of the baseline leaf nucleus/fcmae_vscyto3d_scratch/ipsc_confocal/train.yml.
+# Recipe delta vs that leaf: the segcons_ctrl fork + model seg_consistency (SegConsistencyLoss, Cellpose-SAM cpsam, slab halfwidth 1, 240-px crop (one cellpose tile), per-tile activation checkpointing) + seg_consistency_weight (gradient-norm calibrated at the fork checkpoint).
+# Everything else, and the rationale for it, is the baseline leaf's.
+base:
+- ../../../_internal/shared/model/train_sets/ipsc_confocal.yml
+- ../../../_internal/shared/model/targets/nucleus.yml
+- ../../../_internal/shared/model/data_overlays/fcmae_vscyto3d_fit.yml
+- ../../../_internal/shared/model/model_overlays/fcmae_vscyto3d_fit.yml
+- ../../../_internal/shared/model/launcher_profiles/mode_fit.yml
+- ../../../_internal/shared/model/launcher_profiles/hardware_4gpu.yml
+- ../../../_internal/shared/model/launcher_profiles/runtime_shared.yml
+benchmark:
+  task: virtual_staining
+  organelle: nucleus
+  train_set: ipsc_confocal
+  model_name: fcmae_vscyto3d_scratch_segcons
+  experiment_id: nucleus__ipsc_confocal__fcmae_vscyto3d_scratch_segcons
+data:
+  init_args:
+    augmentations:
+    - class_path: viscy_transforms.RandWeightedCropd
+      init_args:
+        keys:
+        - Phase3D
+        - Nuclei
+        w_key: Nuclei
+        spatial_size:
+        - 20
+        - 600
+        - 600
+        num_samples: 4
+trainer:
+  logger:
+    init_args:
+      name: FCMAE_VSCyto3D_Scratch_iPSC_Nucleus_segcons
+      save_dir: /hpc/projects/comp.micro/virtual_staining/models/dynacell/ipsc/nucleus/fcmae_vscyto3d_scratch_segcons
+  callbacks:
+  - class_path: lightning.pytorch.callbacks.LearningRateMonitor
+    init_args:
+      logging_interval: step
+  - class_path: lightning.pytorch.callbacks.ModelCheckpoint
+    init_args:
+      monitor: loss/validate
+      filename: epoch={epoch}-step={step}-loss={loss/validate:.3f}
+      auto_insert_metric_name: false
+      every_n_epochs: 1
+      save_top_k: 5
+      save_last: true
+      dirpath: /hpc/projects/comp.micro/virtual_staining/models/dynacell/ipsc/nucleus/fcmae_vscyto3d_scratch_segcons/checkpoints
+  max_epochs: 20
+launcher:
+  job_name: FCMAE_VSCyto3D_Scratch_Nucleus_segcons
+  run_root: /hpc/projects/comp.micro/virtual_staining/models/dynacell/ipsc/nucleus/fcmae_vscyto3d_scratch_segcons
+model:
+  init_args:
+    ckpt_path: /hpc/projects/comp.micro/virtual_staining/models/dynacell/ipsc/nucleus/fcmae_vscyto3d_scratch_v2/checkpoints/latest-epoch=199-step=62400.ckpt
+    lr: 4.0e-05
+    warmup_steps: 312
+    seg_consistency:
+      class_path: viscy_utils.losses.SegConsistencyLoss
+      init_args:
+        segmenter:
+          class_path: dynacell.cellpose_dense.CellposeDense
+          init_args:
+            model_name: cpsam
+            checkpoint_tiles: 1
+        slab_halfwidth: 1
+        crop_size: 240
+    seg_consistency_weight: 0.032

applications/dynacell/configs/benchmarks/virtual_staining/nucleus/fcmae_vscyto3d_scratch_segcons_ctrl/ipsc_confocal/predict__a549_mantis_denv.yml

@@ -0,0 +1,44 @@
+# GENERATED by applications/dynacell/tools/generate_spotlight_v2_leaves.py -- edit the
+# generator, not this file (plan: ~/.claude/plans/spotlight-v2.md).
+# Spotlight-v2 `segcons_ctrl` arm (predict) of the baseline leaf nucleus/fcmae_vscyto3d_scratch/ipsc_confocal/predict__a549_mantis_denv.yml.
+# Inference is identical to the baseline's; only the checkpoint, store and names differ.
+# Fine-tune arm: predict the end of the fine-tune; submit with `--ckpt last`.
+# Everything else, and the rationale for it, is the baseline leaf's.
+base:
+- ../../../_internal/shared/model/predict_sets/a549_mantis_h2b_denv.yml
+- ../../../_internal/shared/model/targets/nucleus.yml
+- ../../../_internal/shared/model/model_overlays/fcmae_vscyto3d_predict.yml
+- ../../../_internal/shared/model/launcher_profiles/mode_predict.yml
+- ../../../_internal/shared/model/launcher_profiles/hardware_predict_any_gpu.yml
+- ../../../_internal/shared/model/launcher_profiles/runtime_shared.yml
+benchmark:
+  task: virtual_staining
+  organelle: nucleus
+  trained_on: ipsc_confocal
+  predict_set: a549_mantis_h2b_denv
+  model_name: fcmae_vscyto3d_scratch_segcons_ctrl
+  experiment_id: nucleus__ipsc_confocal__fcmae_vscyto3d_scratch_segcons_ctrl__a549_mantis_h2b_denv
+  dataset_ref:
+    target: h2b
+model:
+  init_args:
+    ckpt_path: /hpc/projects/comp.micro/virtual_staining/models/dynacell/ipsc/nucleus/fcmae_vscyto3d_scratch_segcons_ctrl/checkpoints/last.ckpt
+data:
+  init_args:
+    normalizations:
+    - class_path: viscy_transforms.NormalizeSampled
+      init_args:
+        keys:
+        - Phase3D
+        level: fov_statistics
+        subtrahend: mean
+        divisor: std
+    augmentations: []
+trainer:
+  callbacks:
+  - class_path: viscy_utils.callbacks.prediction_writer.HCSPredictionWriter
+    init_args:
+      output_store: /hpc/projects/virtual_staining/training/dynacell/nucleus/fcmae_vscyto3d_scratch_segcons_ctrl/ipsc/a549__denv/prediction.zarr
+launcher:
+  job_name: FCMAE_VSCyto3D_Scratch_PRED_NUCL_ON_A549_DENV_segcons_ctrl
+  run_root: /hpc/projects/virtual_staining/training/dynacell/nucleus/fcmae_vscyto3d_scratch_segcons_ctrl/ipsc/a549__denv

applications/dynacell/configs/benchmarks/virtual_staining/nucleus/fcmae_vscyto3d_scratch_segcons_ctrl/ipsc_confocal/predict__a549_mantis_mock.yml

@@ -0,0 +1,44 @@
+# GENERATED by applications/dynacell/tools/generate_spotlight_v2_leaves.py -- edit the
+# generator, not this file (plan: ~/.claude/plans/spotlight-v2.md).
+# Spotlight-v2 `segcons_ctrl` arm (predict) of the baseline leaf nucleus/fcmae_vscyto3d_scratch/ipsc_confocal/predict__a549_mantis_mock.yml.
+# Inference is identical to the baseline's; only the checkpoint, store and names differ.
+# Fine-tune arm: predict the end of the fine-tune; submit with `--ckpt last`.
+# Everything else, and the rationale for it, is the baseline leaf's.
+base:
+- ../../../_internal/shared/model/predict_sets/a549_mantis_h2b_mock.yml
+- ../../../_internal/shared/model/targets/nucleus.yml
+- ../../../_internal/shared/model/model_overlays/fcmae_vscyto3d_predict.yml
+- ../../../_internal/shared/model/launcher_profiles/mode_predict.yml
+- ../../../_internal/shared/model/launcher_profiles/hardware_predict_any_gpu.yml
+- ../../../_internal/shared/model/launcher_profiles/runtime_shared.yml
+benchmark:
+  task: virtual_staining
+  organelle: nucleus
+  trained_on: ipsc_confocal
+  predict_set: a549_mantis_h2b_mock
+  model_name: fcmae_vscyto3d_scratch_segcons_ctrl
+  experiment_id: nucleus__ipsc_confocal__fcmae_vscyto3d_scratch_segcons_ctrl__a549_mantis_h2b_mock
+  dataset_ref:
+    target: h2b
+model:
+  init_args:
+    ckpt_path: /hpc/projects/comp.micro/virtual_staining/models/dynacell/ipsc/nucleus/fcmae_vscyto3d_scratch_segcons_ctrl/checkpoints/last.ckpt
+data:
+  init_args:
+    normalizations:
+    - class_path: viscy_transforms.NormalizeSampled
+      init_args:
+        keys:
+        - Phase3D
+        level: fov_statistics
+        subtrahend: mean
+        divisor: std
+    augmentations: []
+trainer:
+  callbacks:
+  - class_path: viscy_utils.callbacks.prediction_writer.HCSPredictionWriter
+    init_args:
+      output_store: /hpc/projects/virtual_staining/training/dynacell/nucleus/fcmae_vscyto3d_scratch_segcons_ctrl/ipsc/a549__mock/prediction.zarr
+launcher:
+  job_name: FCMAE_VSCyto3D_Scratch_PRED_NUCL_ON_A549_MOCK_segcons_ctrl
+  run_root: /hpc/projects/virtual_staining/training/dynacell/nucleus/fcmae_vscyto3d_scratch_segcons_ctrl/ipsc/a549__mock

applications/dynacell/configs/benchmarks/virtual_staining/nucleus/fcmae_vscyto3d_scratch_segcons_ctrl/ipsc_confocal/predict__a549_mantis_zikv.yml

@@ -0,0 +1,44 @@
+# GENERATED by applications/dynacell/tools/generate_spotlight_v2_leaves.py -- edit the
+# generator, not this file (plan: ~/.claude/plans/spotlight-v2.md).
+# Spotlight-v2 `segcons_ctrl` arm (predict) of the baseline leaf nucleus/fcmae_vscyto3d_scratch/ipsc_confocal/predict__a549_mantis_zikv.yml.
+# Inference is identical to the baseline's; only the checkpoint, store and names differ.
+# Fine-tune arm: predict the end of the fine-tune; submit with `--ckpt last`.
+# Everything else, and the rationale for it, is the baseline leaf's.
+base:
+- ../../../_internal/shared/model/predict_sets/a549_mantis_h2b_zikv.yml
+- ../../../_internal/shared/model/targets/nucleus.yml
+- ../../../_internal/shared/model/model_overlays/fcmae_vscyto3d_predict.yml
+- ../../../_internal/shared/model/launcher_profiles/mode_predict.yml
+- ../../../_internal/shared/model/launcher_profiles/hardware_predict_any_gpu.yml
+- ../../../_internal/shared/model/launcher_profiles/runtime_shared.yml
+benchmark:
+  task: virtual_staining
+  organelle: nucleus
+  trained_on: ipsc_confocal
+  predict_set: a549_mantis_h2b_zikv
+  model_name: fcmae_vscyto3d_scratch_segcons_ctrl
+  experiment_id: nucleus__ipsc_confocal__fcmae_vscyto3d_scratch_segcons_ctrl__a549_mantis_h2b_zikv
+  dataset_ref:
+    target: h2b
+model:
+  init_args:
+    ckpt_path: /hpc/projects/comp.micro/virtual_staining/models/dynacell/ipsc/nucleus/fcmae_vscyto3d_scratch_segcons_ctrl/checkpoints/last.ckpt
+data:
+  init_args:
+    normalizations:
+    - class_path: viscy_transforms.NormalizeSampled
+      init_args:
+        keys:
+        - Phase3D
+        level: fov_statistics
+        subtrahend: mean
+        divisor: std
+    augmentations: []
+trainer:
+  callbacks:
+  - class_path: viscy_utils.callbacks.prediction_writer.HCSPredictionWriter
+    init_args:
+      output_store: /hpc/projects/virtual_staining/training/dynacell/nucleus/fcmae_vscyto3d_scratch_segcons_ctrl/ipsc/a549__zikv/prediction.zarr
+launcher:
+  job_name: FCMAE_VSCyto3D_Scratch_PRED_NUCL_ON_A549_ZIKV_segcons_ctrl
+  run_root: /hpc/projects/virtual_staining/training/dynacell/nucleus/fcmae_vscyto3d_scratch_segcons_ctrl/ipsc/a549__zikv