[pose-detection]Change enableSmoothing to modelConfig for MoveNet.

pose-detection/src/blazepose_mediapipe/detector.ts

@@ -143,7 +143,7 @@ class BlazePoseMediaPipeDetector implements PoseDetector {
 /**
  * Loads the MediaPipe solution.
  *
- * @param modelConfig ModelConfig dictionary that contains parameters for
+ * @param modelConfig ModelConfig object that contains parameters for
  * the BlazePose loading process. Please find more details of each parameters
  * in the documentation of the `BlazePoseMediaPipeModelConfig` interface.
  */

pose-detection/src/blazepose_mediapipe/types.ts

@@ -31,12 +31,38 @@ export interface BlazePoseModelConfig extends ModelConfig {
 export interface BlazePoseEstimationConfig extends EstimationConfig {}
 
 /**
- * Mediapipe model loading config.
+ * Model parameters for BlazePose MediaPipe runtime
+ *
+ * `runtime`: Must set to be 'mediapipe'.
+ *
+ * `enableSmoothing`: Optional. A boolean indicating whether to use temporal
+ * filter to smooth the predicted keypoints. Defaults to True. The temporal
+ * filter relies on `performance.now()`. You can override this timestamp by
+ * passing in your own timestamp (in milliseconds) as the third parameter in
+ * `estimatePoses`.
+ *
+ * `modelType`: Optional. Possible values: 'lite'|'full'|'heavy'. Defaults to
+ * 'full'. The model accuracy increases from lite to heavy, while the inference
+ * speed decreases and memory footprint increases. The heavy variant is intended
+ * for applications that require high accuracy, while the lite variant is
+ * intended for latency-critical applications. The full variant is a balanced
+ * option.
+ *
+ * `solutionPath`: Optional. The path to where the wasm binary and model files
+ * are located.
  */
 export interface BlazePoseMediaPipeModelConfig extends BlazePoseModelConfig {
   runtime: 'mediapipe';
   solutionPath?: string;
 }
 
+/**
+ * Pose estimation parameters for BlazePose MediaPipe runtime.
+ *
+ * `maxPoses`: Optional. Defaults to 1. BlazePose only supports 1 pose for now.
+ *
+ * `flipHorizontal`: Optional. Default to false. When image data comes from
+ * camera, the result has to flip horizontally.
+ */
 export interface BlazePoseMediaPipeEstimationConfig extends
     BlazePoseEstimationConfig {}

pose-detection/src/blazepose_tfjs/detector.ts

@@ -96,14 +96,8 @@ class BlazePoseTfjsDetector implements PoseDetector {
    * ImageData|HTMLImageElement|HTMLCanvasElement|HTMLVideoElement The input
    * image to feed through the network.
    *
-   * @param config Optional.
-   *       maxPoses: Optional. Max number of poses to estimate.
-   *       When maxPoses = 1, a single pose is detected, it is usually much more
-   *       efficient than maxPoses > 1. When maxPoses > 1, multiple poses are
-   *       detected.
-   *
-   *       flipHorizontal: Optional. Default to false. When image data comes
-   *       from camera, the result has to flip horizontally.
+   * @param estimationConfig Optional. See `BlazePoseTfjsEstimationConfig`
+   *       documentation for detail.
    *
    * @param timestamp Optional. In milliseconds. This is useful when image is
    *     a tensor, which doesn't have timestamp info. Or to override timestamp
@@ -473,11 +467,9 @@ class BlazePoseTfjsDetector implements PoseDetector {
 }
 
 /**
- * Loads the BlazePose model. The model to be loaded is configurable using the
- * config dictionary `BlazePoseTfjsModelConfig`. Please find more details in
- * the documentation of the `BlazePoseTfjsModelConfig`.
+ * Loads the BlazePose model.
  *
- * @param modelConfig ModelConfig dictionary that contains parameters for
+ * @param modelConfig ModelConfig object that contains parameters for
  * the BlazePose loading process. Please find more details of each parameters
  * in the documentation of the `BlazePoseTfjsModelConfig` interface.
  */

pose-detection/src/blazepose_tfjs/types.ts

@@ -18,7 +18,24 @@
 import {BlazePoseEstimationConfig, BlazePoseModelConfig} from '../blazepose_mediapipe/types';
 
 /**
- * Additional model parameters for BlazePose Tfjs backend
+ * Model parameters for BlazePose TFJS runtime.
+ *
+ * `runtime`: Must set to be 'tfjs'.
+ *
+ * `enableSmoothing`: Optional. A boolean indicating whether to use temporal
+ * filter to smooth the predicted keypoints. Defaults to True. The temporal
+ * filter relies on the currentTime field of the HTMLVideoElement. You can
+ * override this timestamp by passing in your own timestamp (in milliseconds)
+ * as the third parameter in `estimatePoses`. This is useful when the input is
+ * a tensor, which doesn't have the currentTime field. Or in testing, to
+ * simulate different FPS.
+ *
+ * `modelType`: Optional. Possible values: 'lite'|'full'|'heavy'. Defaults to
+ * 'full'. The model accuracy increases from lite to heavy, while the inference
+ * speed decreases and memory footprint increases. The heavy variant is intended
+ * for applications that require high accuracy, while the lite variant is
+ * intended for latency-critical applications. The full variant is a balanced
+ * option.
  *
  * `detectorModelUrl`: Optional. An optional string that specifies custom url of
  * the detector model. This is useful for area/countries that don't have access
@@ -34,5 +51,13 @@ export interface BlazePoseTfjsModelConfig extends BlazePoseModelConfig {
   landmarkModelUrl?: string;
 }
 
+/**
+ * Pose estimation parameters for BlazePose TFJS runtime.
+ *
+ * `maxPoses`: Optional. Defaults to 1. BlazePose only supports 1 pose for now.
+ *
+ * `flipHorizontal`: Optional. Default to false. When image data comes from
+ * camera, the result has to flip horizontally.
+ */
 export interface BlazePoseTfjsEstimationConfig extends
     BlazePoseEstimationConfig {}

pose-detection/src/movenet/constants.ts

@@ -32,12 +32,12 @@ export const MOVENET_SINGLEPOSE_THUNDER_RESOLUTION = 256;
 
 // The default configuration for loading MoveNet.
 export const MOVENET_CONFIG: MoveNetModelConfig = {
-  modelType: SINGLEPOSE_LIGHTNING
+  modelType: SINGLEPOSE_LIGHTNING,
+  enableSmoothing: true
 };
 
 export const MOVENET_SINGLE_POSE_ESTIMATION_CONFIG: MoveNetEstimationConfig = {
-  maxPoses: 1,
-  enableSmoothing: true
+  maxPoses: 1
 };
 
 export const KEYPOINT_FILTER_CONFIG = {

pose-detection/src/movenet/detector.ts

@@ -41,6 +41,7 @@ class MoveNetDetector implements PoseDetector {
       InputResolution = {height: 0, width: 0};
   private readonly keypointIndexByName =
       getKeypointIndexByName(SupportedModels.MoveNet);
+  private readonly enableSmoothing: boolean;
 
   // Global states.
   private keypointsFilter = new KeypointsOneEuroFilter(KEYPOINT_FILTER_CONFIG);
@@ -61,6 +62,7 @@ class MoveNetDetector implements PoseDetector {
       this.modelInputResolution.width = MOVENET_SINGLEPOSE_THUNDER_RESOLUTION;
       this.modelInputResolution.height = MOVENET_SINGLEPOSE_THUNDER_RESOLUTION;
     }
+    this.enableSmoothing = config.enableSmoothing;
   }
 
   /**
@@ -135,8 +137,6 @@ class MoveNetDetector implements PoseDetector {
    * @param config Optional. A configuration object with the following
    * properties:
    *  `maxPoses`: Optional. Has to be set to 1.
-   *  `enableSmoothing`: Optional. Defaults to `true`. When enabled, a temporal
-   *  smoothing filter will be used on the keypoint locations to reduce jitter.
    *
    * @param timestamp Optional. In milliseconds. This is useful when image is
    *     a tensor, which doesn't have timestamp info. Or to override timestamp
@@ -214,7 +214,7 @@ class MoveNetDetector implements PoseDetector {
 
     // Apply the sequential filter before estimating the cropping area to make
     // it more stable.
-    if (timestamp != null && estimationConfig.enableSmoothing) {
+    if (timestamp != null && this.enableSmoothing) {
       keypoints = this.keypointsFilter.apply(keypoints, timestamp);
     }
 

pose-detection/src/movenet/detector_utils.ts

@@ -30,6 +30,10 @@ export function validateModelConfig(modelConfig: MoveNetModelConfig):
         `Should be one of ${VALID_MODELS}`);
   }
 
+  if (config.enableSmoothing == null) {
+    config.enableSmoothing = true;
+  }
+
   return config;
 }
 
@@ -47,9 +51,5 @@ export function validateEstimationConfig(
     throw new Error(`Invalid maxPoses ${config.maxPoses}. Should be 1.`);
   }
 
-  if (config.enableSmoothing == null) {
-    config.enableSmoothing = true;
-  }
-
   return config;
 }

pose-detection/src/movenet/types.ts

@@ -17,30 +17,37 @@
 import {EstimationConfig, ModelConfig} from '../types';
 
 /**
- * Additional MoveNet model loading config.
+ * MoveNet model loading config.
  *
- * 'modelType': Optional. The type of MoveNet model to load, Lighting or
+ * `enableSmoothing`: Optional. A boolean indicating whether to use temporal
+ * filter to smooth the predicted keypoints. Defaults to True. The temporal
+ * filter relies on the currentTime field of the HTMLVideoElement. You can
+ * override this timestamp by passing in your own timestamp (in milliseconds)
+ * as the third parameter in `estimatePoses`. This is useful when the input is
+ * a tensor, which doesn't have the currentTime field. Or in testing, to
+ * simulate different FPS.
+ *
+ * `modelType`: Optional. The type of MoveNet model to load, Lighting or
  * Thunder. Defaults to Lightning. Lightning is a lower capacity model that can
  * run >50FPS on most modern laptops while achieving good performance. Thunder
- * is A higher capacity model that performs better prediction quality while
+ * is a higher capacity model that performs better prediction quality while
  * still achieving real-time (>30FPS) speed. Thunder will lag behind the
  * lightning, but it will pack a punch.
  *
  * `modelUrl`: Optional. An optional string that specifies custom url of the
  * model. This is useful for area/countries that don't have access to the model
- * hosted on TF Hub.
+ * hosted on TF Hub. If not provided, it will load the model specified by
+ * `modelType` from tf.hub.
  */
 export interface MoveNetModelConfig extends ModelConfig {
+  enableSmoothing?: boolean;
   modelType?: string;
   modelUrl?: string;
 }
 
 /**
  * MoveNet Specific Inference Config.
  *
- * `enableSmoothing`: Optional. Defaults to 'true'. When enabled, a temporal
- * smoothing filter will be used on the keypoint locations to reduce jitter.
+ * `maxPoses`: Optional. Defaults to 1. MoveNet only supports 1 pose for now.
  */
-export interface MoveNetEstimationConfig extends EstimationConfig {
-  enableSmoothing?: boolean;
-}
+export interface MoveNetEstimationConfig extends EstimationConfig {}

pose-detection/yarn.lock

@@ -880,7 +880,7 @@
   resolved "https://registry.npmjs.org/@istanbuljs/schema/-/schema-0.1.3.tgz#e45e384e4b8ec16bce2fd903af78450f6bf7ec98"
   integrity sha512-ZXRY4jNvVgSVQ8DL3LTcakaAtXwTVUxE81hslsyD2AtoXW/wVob10HkOJ1X/pAlcI7D+2YoZKg5do8G/w6RYgA==
 
-"@mediapipe/pose@0.3.x":
+"@mediapipe/pose@~0.3.0":
   version "0.3.1620676110"
   resolved "https://registry.npmjs.org/@mediapipe/pose/-/pose-0.3.1620676110.tgz#35d522196db581e213259fbc42290a577bd239ed"
   integrity sha512-macv7l2fuaqLArYvUSuoUrxoOVut4KEFHEqPU0i6SLpfohJFhBRgpJXiy8w+KTWkdh0e7cTLkPTESiV2+u9tNA==