feat(dynacell): Spotlight v2 topology losses (SAUNA-map Dice, soft-clDice) and ER/mito arms

applications/dynacell/configs/benchmarks/virtual_staining/er/fnet3d_vscyto3daug/ipsc_confocal/predict__a549_mantis_denv.yml

@@ -0,0 +1,42 @@
+# FNet3DVSAug predict template (er, SEC61B): trained on iPSC, predicting a549_mantis_denv.
+# ckpt_path is a placeholder; the Spotlight-v2 arms predict with --ckpt best.
+base:
+  - ../../../_internal/shared/model/predict_sets/a549_mantis_sec61b_denv.yml
+  - ../../../_internal/shared/model/targets/er_sec61b.yml
+  - ../../../_internal/shared/model/model_overlays/fnet3d_paper_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: er
+  trained_on: ipsc_confocal
+  predict_set: a549_mantis_sec61b_denv
+  model_name: fnet3d_vscyto3daug
+  experiment_id: er__ipsc_confocal__fnet3d_vscyto3daug__a549_mantis_sec61b_denv
+
+model:
+  init_args:
+    ckpt_path: /hpc/projects/comp.micro/virtual_staining/models/dynacell/ipsc/er/fnet3d_vscyto3daug/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/er/fnet3d_vscyto3daug/ipsc/a549__denv/prediction.zarr
+
+launcher:
+  job_name: FNet3DVSAug_PRED_SEC61B_ON_A549_sec61b_denv
+  run_root: /hpc/projects/virtual_staining/training/dynacell/er/fnet3d_vscyto3daug/ipsc/a549__denv

applications/dynacell/configs/benchmarks/virtual_staining/er/fnet3d_vscyto3daug/ipsc_confocal/predict__a549_mantis_mock.yml

@@ -0,0 +1,42 @@
+# FNet3DVSAug predict template (er, SEC61B): trained on iPSC, predicting a549_mantis_mock.
+# ckpt_path is a placeholder; the Spotlight-v2 arms predict with --ckpt best.
+base:
+  - ../../../_internal/shared/model/predict_sets/a549_mantis_sec61b_mock.yml
+  - ../../../_internal/shared/model/targets/er_sec61b.yml
+  - ../../../_internal/shared/model/model_overlays/fnet3d_paper_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: er
+  trained_on: ipsc_confocal
+  predict_set: a549_mantis_sec61b_mock
+  model_name: fnet3d_vscyto3daug
+  experiment_id: er__ipsc_confocal__fnet3d_vscyto3daug__a549_mantis_sec61b_mock
+
+model:
+  init_args:
+    ckpt_path: /hpc/projects/comp.micro/virtual_staining/models/dynacell/ipsc/er/fnet3d_vscyto3daug/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/er/fnet3d_vscyto3daug/ipsc/a549__mock/prediction.zarr
+
+launcher:
+  job_name: FNet3DVSAug_PRED_SEC61B_ON_A549_sec61b_mock
+  run_root: /hpc/projects/virtual_staining/training/dynacell/er/fnet3d_vscyto3daug/ipsc/a549__mock

applications/dynacell/configs/benchmarks/virtual_staining/er/fnet3d_vscyto3daug/ipsc_confocal/predict__a549_mantis_zikv.yml

@@ -0,0 +1,42 @@
+# FNet3DVSAug predict template (er, SEC61B): trained on iPSC, predicting a549_mantis_zikv.
+# ckpt_path is a placeholder; the Spotlight-v2 arms predict with --ckpt best.
+base:
+  - ../../../_internal/shared/model/predict_sets/a549_mantis_sec61b_zikv.yml
+  - ../../../_internal/shared/model/targets/er_sec61b.yml
+  - ../../../_internal/shared/model/model_overlays/fnet3d_paper_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: er
+  trained_on: ipsc_confocal
+  predict_set: a549_mantis_sec61b_zikv
+  model_name: fnet3d_vscyto3daug
+  experiment_id: er__ipsc_confocal__fnet3d_vscyto3daug__a549_mantis_sec61b_zikv
+
+model:
+  init_args:
+    ckpt_path: /hpc/projects/comp.micro/virtual_staining/models/dynacell/ipsc/er/fnet3d_vscyto3daug/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/er/fnet3d_vscyto3daug/ipsc/a549__zikv/prediction.zarr
+
+launcher:
+  job_name: FNet3DVSAug_PRED_SEC61B_ON_A549_sec61b_zikv
+  run_root: /hpc/projects/virtual_staining/training/dynacell/er/fnet3d_vscyto3daug/ipsc/a549__zikv

applications/dynacell/configs/benchmarks/virtual_staining/er/fnet3d_vscyto3daug/ipsc_confocal/predict__ipsc_confocal.yml

@@ -0,0 +1,42 @@
+# FNet3DVSAug predict template (er, SEC61B): trained on iPSC, predicting ipsc_confocal.
+# ckpt_path is a placeholder; the Spotlight-v2 arms predict with --ckpt best.
+base:
+  - ../../../_internal/shared/model/predict_sets/ipsc_confocal.yml
+  - ../../../_internal/shared/model/targets/er_sec61b.yml
+  - ../../../_internal/shared/model/model_overlays/fnet3d_paper_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: er
+  trained_on: ipsc_confocal
+  predict_set: ipsc_confocal
+  model_name: fnet3d_vscyto3daug
+  experiment_id: er__ipsc_confocal__fnet3d_vscyto3daug__ipsc_confocal
+
+model:
+  init_args:
+    ckpt_path: /hpc/projects/comp.micro/virtual_staining/models/dynacell/ipsc/er/fnet3d_vscyto3daug/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/er/fnet3d_vscyto3daug/ipsc/ipsc/prediction.zarr
+
+launcher:
+  job_name: FNet3DVSAug_PRED_SEC61B
+  run_root: /hpc/projects/virtual_staining/training/dynacell/er/fnet3d_vscyto3daug/ipsc/ipsc

applications/dynacell/configs/benchmarks/virtual_staining/er/fnet3d_vscyto3daug/ipsc_confocal/train.yml

@@ -0,0 +1,56 @@
+# FNet3D "same-augmentations" recipe (Phase 15 Arm B) on er (SEC61B marker) — AICS iPSC confocal.
+# Template baseline for the Spotlight-v2 Stage 2 #5 arms (generate_spotlight_v2_leaves.py emits
+# fnet3d_vscyto3daug_{v2,v2_seed1,segaux,segaux_sauna,segaux_cldice} from it); never run as-is.
+# Same recipe as nucleus/fnet3d_vscyto3daug/ipsc_confocal/train.yml; the target channel is
+# Structure, so the data overlay's Structure-keyed norm/aug/val lists apply unchanged.
+base:
+  - ../../../_internal/shared/model/train_sets/ipsc_confocal.yml
+  - ../../../_internal/shared/model/targets/er_sec61b.yml
+  - ../../../_internal/shared/model/data_overlays/fnet3d_vscyto3daug_fit.yml
+  - ../../../_internal/shared/model/model_overlays/fnet3d_paper_fit.yml
+  - ../../../_internal/shared/model/launcher_profiles/mode_fit.yml
+  - ../../../_internal/shared/model/launcher_profiles/hardware_gpu_any_long.yml
+  - ../../../_internal/shared/model/launcher_profiles/runtime_shared.yml
+
+benchmark:
+  task: virtual_staining
+  organelle: er
+  train_set: ipsc_confocal
+  model_name: fnet3d_vscyto3daug
+  experiment_id: er__ipsc_confocal__fnet3d_vscyto3daug
+
+trainer:
+  # Switch off the fnet3d_paper 32-true default: bf16-mixed halves the
+  # 384x384x32 activation footprint. MSE loss is immune to the bf16-SSIM
+  # cancellation issue. Must match fnet3d_bigpatch (Arm A) for a clean
+  # augmentation-only comparison.
+  precision: bf16-mixed
+  logger:
+    init_args:
+      name: FNet3D_VSAug_iPSC_SEC61B
+      save_dir: /hpc/projects/comp.micro/virtual_staining/models/dynacell/ipsc/er/fnet3d_vscyto3daug
+  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: 4
+        save_last: true
+        dirpath: /hpc/projects/comp.micro/virtual_staining/models/dynacell/ipsc/er/fnet3d_vscyto3daug/checkpoints
+
+launcher:
+  job_name: FNet3DVSAug_iPSC_SEC61B
+  run_root: /hpc/projects/comp.micro/virtual_staining/models/dynacell/ipsc/er/fnet3d_vscyto3daug
+  # 384x384x32 at bf16-mixed measured ~41GB GPU on the bigpatch arm, so an
+  # 80GB card (H100) fits with room; allow h100|h200 to schedule sooner. The
+  # vscyto3d aug stack (affine grid_sample + intensity) runs somewhat higher
+  # but still well under 80GB. SEC61B.zarr-backed preload pushes MaxVMSize past
+  # the shared 256G cap; mem 512G matches the fnet3d iPSC headroom.
+  sbatch:
+    constraint: "h100|h200"
+    mem: "512G"

applications/dynacell/configs/benchmarks/virtual_staining/er/fnet3d_vscyto3daug_segaux/ipsc_confocal/predict__a549_mantis_denv.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 `segaux` arm (predict) of the baseline leaf er/fnet3d_vscyto3daug/ipsc_confocal/predict__a549_mantis_denv.yml.
+# Inference is identical to the baseline's; only the checkpoint, store and names differ.
+# ckpt_path is a PLACEHOLDER (last.ckpt): submit with `--ckpt best`, re-bake afterwards.
+# Everything else, and the rationale for it, is the baseline leaf's.
+base:
+- ../../../_internal/shared/model/predict_sets/a549_mantis_sec61b_denv.yml
+- ../../../_internal/shared/model/targets/er_sec61b.yml
+- ../../../_internal/shared/model/model_overlays/fnet3d_paper_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: er
+  trained_on: ipsc_confocal
+  predict_set: a549_mantis_sec61b_denv
+  model_name: fnet3d_vscyto3daug_segaux
+  experiment_id: er__ipsc_confocal__fnet3d_vscyto3daug_segaux__a549_mantis_sec61b_denv
+model:
+  init_args:
+    ckpt_path: /hpc/projects/comp.micro/virtual_staining/models/dynacell/ipsc/er/fnet3d_vscyto3daug_segaux/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/er/fnet3d_vscyto3daug_segaux/ipsc/a549__denv/prediction.zarr
+launcher:
+  job_name: FNet3DVSAug_PRED_SEC61B_ON_A549_sec61b_denv_segaux
+  run_root: /hpc/projects/virtual_staining/training/dynacell/er/fnet3d_vscyto3daug_segaux/ipsc/a549__denv

applications/dynacell/configs/benchmarks/virtual_staining/er/fnet3d_vscyto3daug_segaux/ipsc_confocal/predict__a549_mantis_mock.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 `segaux` arm (predict) of the baseline leaf er/fnet3d_vscyto3daug/ipsc_confocal/predict__a549_mantis_mock.yml.
+# Inference is identical to the baseline's; only the checkpoint, store and names differ.
+# ckpt_path is a PLACEHOLDER (last.ckpt): submit with `--ckpt best`, re-bake afterwards.
+# Everything else, and the rationale for it, is the baseline leaf's.
+base:
+- ../../../_internal/shared/model/predict_sets/a549_mantis_sec61b_mock.yml
+- ../../../_internal/shared/model/targets/er_sec61b.yml
+- ../../../_internal/shared/model/model_overlays/fnet3d_paper_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: er
+  trained_on: ipsc_confocal
+  predict_set: a549_mantis_sec61b_mock
+  model_name: fnet3d_vscyto3daug_segaux
+  experiment_id: er__ipsc_confocal__fnet3d_vscyto3daug_segaux__a549_mantis_sec61b_mock
+model:
+  init_args:
+    ckpt_path: /hpc/projects/comp.micro/virtual_staining/models/dynacell/ipsc/er/fnet3d_vscyto3daug_segaux/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/er/fnet3d_vscyto3daug_segaux/ipsc/a549__mock/prediction.zarr
+launcher:
+  job_name: FNet3DVSAug_PRED_SEC61B_ON_A549_sec61b_mock_segaux
+  run_root: /hpc/projects/virtual_staining/training/dynacell/er/fnet3d_vscyto3daug_segaux/ipsc/a549__mock

applications/dynacell/configs/benchmarks/virtual_staining/er/fnet3d_vscyto3daug_segaux/ipsc_confocal/predict__a549_mantis_zikv.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 `segaux` arm (predict) of the baseline leaf er/fnet3d_vscyto3daug/ipsc_confocal/predict__a549_mantis_zikv.yml.
+# Inference is identical to the baseline's; only the checkpoint, store and names differ.
+# ckpt_path is a PLACEHOLDER (last.ckpt): submit with `--ckpt best`, re-bake afterwards.
+# Everything else, and the rationale for it, is the baseline leaf's.
+base:
+- ../../../_internal/shared/model/predict_sets/a549_mantis_sec61b_zikv.yml
+- ../../../_internal/shared/model/targets/er_sec61b.yml
+- ../../../_internal/shared/model/model_overlays/fnet3d_paper_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: er
+  trained_on: ipsc_confocal
+  predict_set: a549_mantis_sec61b_zikv
+  model_name: fnet3d_vscyto3daug_segaux
+  experiment_id: er__ipsc_confocal__fnet3d_vscyto3daug_segaux__a549_mantis_sec61b_zikv
+model:
+  init_args:
+    ckpt_path: /hpc/projects/comp.micro/virtual_staining/models/dynacell/ipsc/er/fnet3d_vscyto3daug_segaux/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/er/fnet3d_vscyto3daug_segaux/ipsc/a549__zikv/prediction.zarr
+launcher:
+  job_name: FNet3DVSAug_PRED_SEC61B_ON_A549_sec61b_zikv_segaux
+  run_root: /hpc/projects/virtual_staining/training/dynacell/er/fnet3d_vscyto3daug_segaux/ipsc/a549__zikv