[opt] Make Stardist code DRY

- create abstract stardist class for the two concrete stardist variant
  classes, because most of the functionality is identical

src/utils/models/StardistFluo/StardistFluo.ts → src/utils/models/Stardist/AbstractStardist.ts

@@ -1,23 +1,16 @@
 import { GraphModel, History, LayersModel } from "@tensorflow/tfjs";
 
 import { Segmenter } from "../AbstractSegmenter/AbstractSegmenter";
-import { loadStardist } from "./loadStardist";
 import { preprocessStardist } from "./preprocessStardist";
 import { predictStardist } from "./predictStardist";
 import { generateUUID } from "utils/common/helpers";
 import { LoadInferenceDataArgs } from "../types";
-import { ModelTask } from "../enums";
 import { Kind, ImageObject } from "store/data/types";
 
 /*
- * Stardist (Versatile) Fluorescence Nuclei Segmentation
- * https://zenodo.org/records/6348085
- * https://bioimage.io/#/?tags=stardist&id=10.5281%2Fzenodo.6348084
- * https://github.com/stardist/stardist/blob/master/README.md#pretrained-models-for-2d
- * Stardist: model for object detection / instance segmentation with star-convex shapes
- * This pretrained model: meant to segment individual cell nuclei from single channel fluorescence data (2018 DSB)
+ * Abstract model for Stardist variants
  */
-export class StardistFluo extends Segmenter {
+export abstract class Stardist extends Segmenter {
   protected _fgKind?: Kind;
   protected _inferenceDataDims?: Array<{
     width: number;
@@ -26,25 +19,7 @@ export class StardistFluo extends Segmenter {
     padY: number;
   }>;
 
-  constructor() {
-    super({
-      name: "StardistFluo",
-      task: ModelTask.Segmentation,
-      graph: true,
-      pretrained: true,
-      trainable: false,
-      requiredChannels: 3,
-    });
-  }
-
-  public async loadModel() {
-    if (this._model) return;
-    // inputs: [ {name: 'input', shape: [-1,-1,-1,1], dtype: 'float32'} ]
-    // outputs: [ {name: 'concatenate_4/concat', shape: [-1, -1, -1, 33], dtype: 'float32'} ]
-    // where each -1 matches on input and output of corresponding dim/axis
-    // 33 -> 1 probability score, followed by 32 radial equiangular distances of rays
-    this._model = await loadStardist();
-  }
+  public abstract loadModel(): Promise<void>;
 
   public loadTraining(images: ImageObject[], preprocessingArgs: any): void {}
 

src/utils/models/Stardist/StardistFluo/StardistFluo.ts

@@ -0,0 +1,33 @@
+import { Stardist } from "../AbstractStardist";
+import { loadStardistFluo } from "./loadStardistFluo";
+import { ModelTask } from "../../enums";
+
+/*
+ * Stardist (Versatile) Fluorescence Nuclei Segmentation
+ * https://zenodo.org/records/6348085
+ * https://bioimage.io/#/?tags=stardist&id=10.5281%2Fzenodo.6348084
+ * https://github.com/stardist/stardist/blob/master/README.md#pretrained-models-for-2d
+ * Stardist: model for object detection / instance segmentation with star-convex shapes
+ * This pretrained model: meant to segment individual cell nuclei from single channel fluorescence data (2018 DSB)
+ */
+export class StardistFluo extends Stardist {
+  constructor() {
+    super({
+      name: "StardistFluo",
+      task: ModelTask.Segmentation,
+      graph: true,
+      pretrained: true,
+      trainable: false,
+      requiredChannels: 3,
+    });
+  }
+
+  public async loadModel() {
+    if (this._model) return;
+    // inputs: [ {name: 'input', shape: [-1,-1,-1,1], dtype: 'float32'} ]
+    // outputs: [ {name: 'concatenate_4/concat', shape: [-1, -1, -1, 33], dtype: 'float32'} ]
+    // where each -1 matches on input and output of corresponding dim/axis
+    // 33 -> 1 probability score, followed by 32 radial equiangular distances of rays
+    this._model = await loadStardistFluo();
+  }
+}

src/utils/models/StardistFluo/loadStardist.ts → src/utils/models/Stardist/StardistFluo/loadStardistFluo.ts

@@ -19,7 +19,7 @@ import Stardist2DFluorescenceWeights2 from "data/model-data/stardist-fluo/group1
  *   from relative paths described by the paths fields in weights manifest.
  */
 
-export async function loadStardist() {
+export async function loadStardistFluo() {
   let modelDescription: File;
   let modelWeights1: File;
   let modelWeights2: File;
@@ -50,7 +50,7 @@ export async function loadStardist() {
     const error: Error = err as Error;
     process.env.NODE_ENV !== "production" &&
       process.env.REACT_APP_LOG_LEVEL === "1" &&
-      console.error(`error loading stardist: ${error.message}`);
+      console.error(`error loading stardist fluorescence: ${error.message}`);
     throw err;
   }
 
@@ -65,7 +65,7 @@ export async function loadStardist() {
 
     process.env.NODE_ENV !== "production" &&
       process.env.REACT_APP_LOG_LEVEL === "1" &&
-      console.error(`error loading stardist: ${error.message}`);
+      console.error(`error loading stardist fluorescence: ${error.message}`);
 
     throw err;
   }

src/utils/models/Stardist/StardistVHE/StardistVHE.ts

@@ -0,0 +1,33 @@
+import { Stardist } from "../AbstractStardist";
+import { loadStardistVHE } from "./loadStardistVHE";
+import { ModelTask } from "../../enums";
+
+/*
+ * Stardist (Versatile) H&E Nuclei Segmentation
+ * https://zenodo.org/record/6338615
+ * https://bioimage.io/#/?tags=stardist&id=10.5281%2Fzenodo.6338614&type=model
+ * https://github.com/stardist/stardist/blob/master/README.md#pretrained-models-for-2d
+ * Stardist: model for object detection / instance segmentation with star-convex shapes
+ * This pretrained model: meant to segment individual cell nuclei from brightfield images with H&E staining
+ */
+export class StardistVHE extends Stardist {
+  constructor() {
+    super({
+      name: "StardistVHE",
+      task: ModelTask.Segmentation,
+      graph: true,
+      pretrained: true,
+      trainable: false,
+      requiredChannels: 3,
+    });
+  }
+
+  public async loadModel() {
+    if (this._model) return;
+    // inputs: [ {name: 'input', shape: [-1,-1,-1,3], dtype: 'float32'} ]
+    // outputs: [ {name: 'concatenate_4/concat', shape: [-1, -1, -1, 33], dtype: 'float32'} ]
+    // where each -1 matches on input and output of corresponding dim/axis
+    // 33 -> 1 probability score, followed by 32 radial equiangular distances of rays
+    this._model = await loadStardistVHE();
+  }
+}

src/utils/models/StardistVHE/loadStardist.ts → src/utils/models/Stardist/StardistVHE/loadStardistVHE.ts

@@ -19,7 +19,7 @@ import Stardist2DBrightfieldWeights2 from "data/model-data//stardist-vhe/group1-
  *   from relative paths described by the paths fields in weights manifest.
  */
 
-export async function loadStardist() {
+export async function loadStardistVHE() {
   let modelDescription: File;
   let modelWeights1: File;
   let modelWeights2: File;
@@ -50,7 +50,7 @@ export async function loadStardist() {
     const error: Error = err as Error;
     process.env.NODE_ENV !== "production" &&
       process.env.REACT_APP_LOG_LEVEL === "1" &&
-      console.error(`error loading stardist: ${error.message}`);
+      console.error(`error loading stardist H&E: ${error.message}`);
     throw err;
   }
 
@@ -65,7 +65,7 @@ export async function loadStardist() {
 
     process.env.NODE_ENV !== "production" &&
       process.env.REACT_APP_LOG_LEVEL === "1" &&
-      console.error(`error loading stardist: ${error.message}`);
+      console.error(`error loading stardist H&E: ${error.message}`);
 
     throw err;
   }

src/utils/models/Stardist/index.ts

@@ -0,0 +1,2 @@
+export { StardistFluo } from "./StardistFluo";
+export { StardistVHE } from "./StardistVHE";