Add object detection and image segmentation models to the task API

[jinjingforever]

• The code for object detection and image segmentation is very similar structurally to the image classification code.
• Add the summary table to README. See preview here. The API doc links are placeholders. The other links (demos, model requirements, model collections) should work. I will add more stuff to this table as I add more models to the task API.
• Minor tweaks to code to make things better for doc-gen and usability.

---

This change is 

[pyu10055]

Reviewable status: complete! 1 of 1 approvals obtained (waiting on @jinjingforever)

---

coco-ssd/src/index.ts, line 42 at r1 (raw file):

export interface ModelConfig {
  /**
   * It determines wich PoseNet architecture to load. The supported

typo PoseNet: object detection

---

tasks/package.json, line 30 at r1 (raw file):

    "@tensorflow/tfjs-core": "^3.5.0",
    "@tensorflow/tfjs-tflite": "0.0.1-alpha.3",
    "@tensorflow-models/mobilenet": "link:../mobilenet",

should this be updated to a fix version before releasing?

---

tasks/src/tasks/image_segmentation/deeplab_tfjs.ts, line 67 at r1 (raw file):

      options.quantizationBytes = 2;
    }
    const deeplabModel = await sourceModelGlobal.load(options);

can you remind me how the dependencies are loaded for the model npm?

---

tasks/src/tasks/object_detection/tflite_common.ts, line 50 at r1 (raw file):

    });
    const finalResult: ObjectDetectionResult = {
      objects,

remove the last ,

[jinjingforever]

Reviewable status: complete! 1 of 1 approvals obtained (waiting on @pyu10055)

---

coco-ssd/src/index.ts, line 42 at r1 (raw file):

Previously, pyu10055 (Ping Yu) wrote…

typo PoseNet: object detection

Done.

---

tasks/package.json, line 30 at r1 (raw file):

Previously, pyu10055 (Ping Yu) wrote…

should this be updated to a fix version before releasing?

I think this part is probably fine. This is only for development. The actual versions of package to load are in the task model's metadata.

---

tasks/src/tasks/image_segmentation/deeplab_tfjs.ts, line 67 at r1 (raw file):

Previously, pyu10055 (Ping Yu) wrote…

can you remind me how the dependencies are loaded for the model npm?

The package loader will load the packages defined in packageUrls. When it finishes, the global namespace (e.g. mobilenet, deeplab, etc) will be initialized by loaded packages. That global namespace variable will be passed to this transformSourceModel method (the sourceModelGlobal parameter) and it will be used here to call its function (load in this case) to load the source model.

---

tasks/src/tasks/object_detection/tflite_common.ts, line 50 at r1 (raw file):

Previously, pyu10055 (Ping Yu) wrote…

remove the last ,

Done.