feat(virtual-backgrounds) use new Open Source model

https://google.github.io/mediapipe/solutions/models.html#selfie-segmentation

react/features/stream-effects/virtual-background/JitsiStreamBackgroundEffect.js

         const outputMemoryOffset = this._model._getOutputMemoryOffset() / 4;
 
         for (let i = 0; i < this._segmentationPixelCount; i++) {
-            const background = this._model.HEAPF32[outputMemoryOffset + (i * 2)];
-            const person = this._model.HEAPF32[outputMemoryOffset + (i * 2) + 1];
-            const shift = Math.max(background, person);
-            const backgroundExp = Math.exp(background - shift);
-            const personExp = Math.exp(person - shift);
+            const person = this._model.HEAPF32[outputMemoryOffset + i];
 
             // Sets only the alpha component of each pixel.
-            this._segmentationMask.data[(i * 4) + 3] = (255 * personExp) / (backgroundExp + personExp);
+            this._segmentationMask.data[(i * 4) + 3] = 255 * person;
+
         }
         this._segmentationMaskCtx.putImageData(this._segmentationMask, 0, 0);
     }

react/features/stream-effects/virtual-background/index.js

 import createTFLiteModule from './vendor/tflite/tflite';
 import createTFLiteSIMDModule from './vendor/tflite/tflite-simd';
 const models = {
-    model96: 'libs/segm_lite_v681.tflite',
-    model144: 'libs/segm_full_v679.tflite'
+    modelLandscape: 'libs/selfie_segmentation_landscape.tflite'
 };
 
 let tflite;
 let isWasmDisabled = false;
 
 const segmentationDimensions = {
-    model96: {
-        height: 96,
-        width: 160
-    },
-    model144: {
+    modelLandscape: {
         height: 144,
         width: 256
     }
     }
 
     const modelBufferOffset = tflite._getModelBufferMemoryOffset();
-    const modelResponse = await fetch(wasmCheck.feature.simd ? models.model144 : models.model96);
+    const modelResponse = await fetch(models.modelLandscape);
 
     if (!modelResponse.ok) {
         throw new Error('Failed to download tflite model!');
     tflite._loadModel(model.byteLength);
 
     const options = {
-        ...wasmCheck.feature.simd ? segmentationDimensions.model144 : segmentationDimensions.model96,
+        ...segmentationDimensions.modelLandscape,
         virtualBackground
     };
 

react/features/stream-effects/virtual-background/vendor/README.md

 # Virtual Background on stream effects
 
-> Inspired from https://ai.googleblog.com/2020/10/background-features-in-google-meet.html and https://github.com/Volcomix/virtual-background.git
+> From https://google.github.io/mediapipe/solutions/models.html#selfie-segmentation
 
 #### Canvas 2D + CPU
 
 - [WebAssembly](https://webassembly.org/)
 - [WebAssembly SIMD](https://github.com/WebAssembly/simd)
 - [TFLite](https://blog.tensorflow.org/2020/07/accelerating-tensorflow-lite-xnnpack-integration.html)
-
-## LICENSE
-
-The mdoels vendored here were downloaded early January (they were available as early as the 4th), before Google switched the license away from Apache 2. Thus we understand they are not covered by the new license which according to the [model card](https://drive.google.com/file/d/1lnP1bRi9CSqQQXUHa13159vLELYDgDu0/view) dates from the 21st of January.
-
-We are not lawyers so do get legal advise if in doubt.
-
-References:
-
-- Model license discussion: https://github.com/tensorflow/tfjs/issues/4177
-- Current vendored model is discovered: https://github.com/tensorflow/tfjs/issues/4177#issuecomment-753934631
-- License change is noticed: https://github.com/tensorflow/tfjs/issues/4177#issuecomment-771536641