Add fluorescence illumination-correction and image-restoration workers

.gitignore

@@ -0,0 +1,4 @@
+__pycache__/
+*.py[cod]
+*.egg-info/
+.pytest_cache/

REGISTRY.md

@@ -7,13 +7,13 @@ Auto-generated by `generate_worker_docs.py` -- do not edit manually.
 
 | Category | Count |
 |----------|------:|
-| Annotation Workers | 26 |
+| Annotation Workers | 28 |
 | Property Workers -- Blobs | 10 |
 | Property Workers -- Points | 8 |
 | Property Workers -- Lines | 3 |
 | Property Workers -- Connections | 2 |
 | Test / Sample Workers | 7 |
-| **Total** | **56** |
+| **Total** | **58** |
 
 ## Annotation Workers
 
@@ -36,6 +36,8 @@ Create new annotations by segmenting images or connecting existing annotations.
 | Gaussian Blur | Applies Gaussian blur to images |  | Yes | [docs](workers/annotations/gaussian_blur/GAUSSIAN_BLUR.md) |
 | H&E Deconvolution | Deconvolves H&E stains |  | Yes | [docs](workers/annotations/h_and_e_deconvolution/H_AND_E_DECONVOLUTION.md) |
 | Histogram Matching | Corrects images using histogram matching |  | Yes | [docs](workers/annotations/histogram_matching/HISTOGRAM_MATCHING.md) |
+| Illumination Correction | Corrects illumination/shading/striping (BaSiC, CIDRE, CellProfiler, flat-field, destripe, SSCOR) |  | Yes | [docs](workers/annotations/illumination_correction/ILLUMINATION_CORRECTION.md) |
+| Image Restoration | Restores images (Noise2Void/N2V2, Cellpose3, ZS-DeconvNet, FluoResFM) | Yes | Yes | [docs](workers/annotations/image_restoration/IMAGE_RESTORATION.md) |
 | Laplacian of Gaussian | This tool finds spots in an image using the Laplacian of Gaussian method.It uses a filt... |  |  | [docs](workers/annotations/laplacian_of_gaussian/LAPLACIAN_OF_GAUSSIAN.md) |
 | Time lapse registration | Corrects images using time lapse registration |  | Yes | [docs](workers/annotations/registration/REGISTRATION.md) |
 | Rolling Ball | Corrects images using a rolling ball |  | Yes | [docs](workers/annotations/rolling_ball/ROLLING_BALL.md) |

docker-compose.yml

@@ -224,6 +224,24 @@ services:
     depends_on:
       - image-processing-base
 
+  illumination_correction:
+    build:
+      context: .
+      dockerfile: ./workers/annotations/illumination_correction/Dockerfile
+    image: annotations/illumination_correction:latest
+    profiles: ["worker", "image-processing"]
+    depends_on:
+      - image-processing-base
+
+  image_restoration:
+    build:
+      context: .
+      dockerfile: ./workers/annotations/image_restoration/${DOCKERFILE:-Dockerfile}
+    image: annotations/image_restoration:latest
+    profiles: ["worker", "image-processing"]
+    depends_on:
+      - image-processing-base
+
   # AI workers
   ai_analysis:
     build:
@@ -473,3 +491,21 @@ services:
     depends_on:
       - deconwolf
     profiles: ["test"]
+
+  illumination_correction_test:
+    build:
+      context: .
+      dockerfile: ./workers/annotations/illumination_correction/tests/Dockerfile_Test
+    image: annotations/illumination_correction:test
+    depends_on:
+      - illumination_correction
+    profiles: ["test"]
+
+  image_restoration_test:
+    build:
+      context: .
+      dockerfile: ./workers/annotations/image_restoration/tests/Dockerfile_Test
+    image: annotations/image_restoration:test
+    depends_on:
+      - image_restoration
+    profiles: ["test"]

workers/annotations/illumination_correction/Dockerfile

@@ -0,0 +1,68 @@
+FROM nimbusimage/image-processing-base:latest
+
+# The base image's "worker" conda env already has numpy>=2.0 / scipy /
+# scikit-image and large_image; activate it to install this worker's extra
+# dependencies.
+SHELL ["conda", "run", "-n", "worker", "/bin/bash", "-c"]
+
+# large_image wheels (mirrors deconwolf's pattern; already present in the base
+# image, pinned again here for explicitness/self-containment).
+RUN pip install large-image-source-zarr large-image-source-tiff large-image-converter large-image --find-links https://girder.github.io/large_image_wheels
+
+# CPU-only torch/torchvision. The base is a plain (non-CUDA) image, so pull the
+# CPU wheels explicitly -- the default PyPI torch wheel bundles a multi-GB CUDA
+# runtime that would never be used here. Installing torch first also lets us
+# satisfy basicpy's/SSCOR's torch dependency with the CPU build.
+RUN pip install torch torchvision --index-url https://download.pytorch.org/whl/cpu
+
+# BaSiCPy (BaSiC shading/darkfield correction) for the `basic` method.
+# Installed with --no-deps, then its runtime deps by hand: BaSiCPy 2.0.0
+# hard-pins scipy<1.13, but the base env is numpy>=2.0 (which needs scipy>=1.13,
+# already installed and linked by the conda scikit-image/large_image binaries).
+# Letting pip honor basicpy's pin would downgrade scipy and break those
+# numpy-2-ABI C extensions at runtime, so we keep the base scipy and install
+# only basicpy's *other* runtime deps (torch is the CPU build installed above).
+RUN pip install --no-deps basicpy && \
+    pip install "hyperactive>=4.4.0,<5" torch-dct pooch "pydantic>=2.0.0" dask tqdm
+
+# pystripe (wavelet-FFT destriping) for the `destripe` method.
+RUN pip install pystripe
+
+# --- SSCOR (deep-learning stripe self-correction) ---
+# Vendored from https://github.com/lxxcontinue/SSCOR, pinned to commit
+# 985479cd79bcf1359e3d9ba44bacd5f372eb2e60. It's a pytorch-CycleGAN-and-pix2pix
+# -style codebase with no clean importable inference API, so `correct_sscor` in
+# entrypoint.py shells out to its CLI scripts (restore.py / train.py / sampling;
+# see SSCOR_REPO_PATH).
+RUN git clone https://github.com/lxxcontinue/SSCOR.git /sscor && \
+    cd /sscor && git checkout 985479cd79bcf1359e3d9ba44bacd5f372eb2e60
+
+# Extra deps SSCOR's pix2pix-style pipeline needs (torch/torchvision are the CPU
+# builds installed above).
+RUN pip install dominate visdom opencv-python matplotlib wandb
+
+# NOTE: SSCOR weights are NOT baked into this image -- the upstream repo
+# distributes trained generator checkpoints via Google Drive, not a stable
+# direct-download URL, and producing one requires an image-specific offline
+# self-training stage (see ILLUMINATION_CORRECTION.md). For SSCOR mode
+# "pretrained", supply a checkpoint at runtime by mounting it into the container
+# and setting the SSCOR_WEIGHTS environment variable to its path (staged
+# internally as the `latest_net_G_A.pth` the CycleGAN loader expects); the
+# `sscor` Method sendError()s with instructions if it is unset/missing. SSCOR
+# mode "self-train" needs no checkpoint (it trains one per frame).
+
+# entrypoint.py is copied last so routine edits don't invalidate the (slow)
+# dependency/clone layers above.
+COPY ./workers/annotations/illumination_correction/entrypoint.py /
+
+LABEL isUPennContrastWorker="" \
+      isAnnotationWorker="" \
+      interfaceName="Illumination Correction" \
+      interfaceCategory="Image Processing" \
+      description="Corrects illumination/shading/striping (BaSiC, CIDRE, CellProfiler, flat-field, destripe, SSCOR)" \
+      hasPreview="False" \
+      advancedOptionsPanel="False" \
+      annotationConfigurationPanel="False" \
+      defaultToolName="Illumination Correction"
+
+# ENTRYPOINT inherited from image-processing-base (run_worker.sh /entrypoint.py — see todo/worker-startup-latency.md)