make padToSquare padding order append first when size is uneven + fixed issue…

build/padToSquare.js

 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
 var tf = require("@tensorflow/tfjs-core");
-var utils_1 = require("./utils");
 /**
  * Pads the smaller dimension of an image tensor with zeros, such that width === height.
  *
@@ -17,16 +16,24 @@ function padToSquare(imgTensor, isCenterImage) {
             return imgTensor;
         }
         var dimDiff = Math.abs(height - width);
-        var paddingAmount = Math.floor(dimDiff * (isCenterImage ? 0.5 : 1));
+        var paddingAmount = Math.round(dimDiff * (isCenterImage ? 0.5 : 1));
         var paddingAxis = height > width ? 2 : 1;
-        var getPaddingTensorShape = function (isRoundUp) {
-            if (isRoundUp === void 0) { isRoundUp = false; }
+        var createPaddingTensor = function (paddingAmount) {
             var paddingTensorShape = imgTensor.shape.slice();
-            paddingTensorShape[paddingAxis] = paddingAmount + (isRoundUp ? 1 : 0);
-            return paddingTensorShape;
+            paddingTensorShape[paddingAxis] = paddingAmount;
+            return tf.fill(paddingTensorShape, 0);
         };
-        var tensorsToStack = (isCenterImage ? [tf.fill(getPaddingTensorShape(!utils_1.isEven(dimDiff)), 0)] : [])
-            .concat([imgTensor, tf.fill(getPaddingTensorShape(), 0)]);
+        var paddingTensorAppend = createPaddingTensor(paddingAmount);
+        var remainingPaddingAmount = dimDiff - paddingTensorAppend.shape[paddingAxis];
+        var paddingTensorPrepend = isCenterImage && remainingPaddingAmount
+            ? createPaddingTensor(remainingPaddingAmount)
+            : null;
+        var tensorsToStack = [
+            paddingTensorPrepend,
+            imgTensor,
+            paddingTensorAppend
+        ]
+            .filter(function (t) { return t !== null; });
         return tf.concat(tensorsToStack, paddingAxis);
     });
 }

build/padToSquare.js.map

-{"version":3,"file":"padToSquare.js","sourceRoot":"","sources":["../src/padToSquare.ts"],"names":[],"mappings":";;AAAA,0CAA4C;AAE5C,iCAAiC;AAEjC;;;;;;GAMG;AACH,qBACE,SAAsB,EACtB,aAA8B;IAA9B,8BAAA,EAAA,qBAA8B;IAE9B,OAAO,EAAE,CAAC,IAAI,CAAC;QAEP,IAAA,6BAA0C,EAAzC,cAAM,EAAE,aAAK,CAA4B;QAChD,IAAI,MAAM,KAAK,KAAK,EAAE;YACpB,OAAO,SAAS,CAAA;SACjB;QAED,IAAM,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,MAAM,GAAG,KAAK,CAAC,CAAA;QACxC,IAAM,aAAa,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,GAAG,CAAC,aAAa,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAA;QACrE,IAAM,WAAW,GAAG,MAAM,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAA;QAE1C,IAAM,qBAAqB,GAAG,UAAC,SAA0B;YAA1B,0BAAA,EAAA,iBAA0B;YACvD,IAAM,kBAAkB,GAAG,SAAS,CAAC,KAAK,CAAC,KAAK,EAAE,CAAA;YAClD,kBAAkB,CAAC,WAAW,CAAC,GAAG,aAAa,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAA;YACrE,OAAO,kBAAkB,CAAA;QAC3B,CAAC,CAAA;QAED,IAAM,cAAc,GAAG,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,IAAI,CAAC,qBAAqB,CAAC,CAAC,cAAM,CAAC,OAAO,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;aAChG,MAAM,CAAC,CAAC,SAAS,EAAG,EAAE,CAAC,IAAI,CAAC,qBAAqB,EAAE,EAAE,CAAC,CAAC,CAAC,CAAkB,CAAA;QAC7E,OAAO,EAAE,CAAC,MAAM,CAAC,cAAc,EAAE,WAAW,CAAC,CAAA;IAC/C,CAAC,CAAC,CAAA;AACJ,CAAC;AAzBD,kCAyBC"}
+{"version":3,"file":"padToSquare.js","sourceRoot":"","sources":["../src/padToSquare.ts"],"names":[],"mappings":";;AAAA,0CAA4C;AAI5C;;;;;;GAMG;AACH,qBACE,SAAsB,EACtB,aAA8B;IAA9B,8BAAA,EAAA,qBAA8B;IAE9B,OAAO,EAAE,CAAC,IAAI,CAAC;QAEP,IAAA,6BAA0C,EAAzC,cAAM,EAAE,aAAK,CAA4B;QAChD,IAAI,MAAM,KAAK,KAAK,EAAE;YACpB,OAAO,SAAS,CAAA;SACjB;QAED,IAAM,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,MAAM,GAAG,KAAK,CAAC,CAAA;QACxC,IAAM,aAAa,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,GAAG,CAAC,aAAa,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAA;QACrE,IAAM,WAAW,GAAG,MAAM,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAA;QAE1C,IAAM,mBAAmB,GAAG,UAAC,aAAqB;YAChD,IAAM,kBAAkB,GAAG,SAAS,CAAC,KAAK,CAAC,KAAK,EAAE,CAAA;YAClD,kBAAkB,CAAC,WAAW,CAAC,GAAG,aAAa,CAAA;YAC/C,OAAO,EAAE,CAAC,IAAI,CAAC,kBAAkB,EAAE,CAAC,CAAC,CAAA;QACvC,CAAC,CAAA;QAED,IAAM,mBAAmB,GAAG,mBAAmB,CAAC,aAAa,CAAC,CAAA;QAC9D,IAAM,sBAAsB,GAAG,OAAO,GAAG,mBAAmB,CAAC,KAAK,CAAC,WAAW,CAAC,CAAA;QAE/E,IAAM,oBAAoB,GAAG,aAAa,IAAI,sBAAsB;YAClE,CAAC,CAAC,mBAAmB,CAAC,sBAAsB,CAAC;YAC7C,CAAC,CAAC,IAAI,CAAA;QAER,IAAM,cAAc,GAAG;YACrB,oBAAoB;YACpB,SAAS;YACT,mBAAmB;SACpB;aACE,MAAM,CAAC,UAAA,CAAC,IAAI,OAAA,CAAC,KAAK,IAAI,EAAV,CAAU,CAAkB,CAAA;QAC3C,OAAO,EAAE,CAAC,MAAM,CAAC,cAAc,EAAE,WAAW,CAAC,CAAA;IAC/C,CAAC,CAAC,CAAA;AACJ,CAAC;AApCD,kCAoCC"}
\ No newline at end of file

dist/face-api.js

@@ -785,16 +785,24 @@
               return imgTensor;
           }
           var dimDiff = Math.abs(height - width);
-          var paddingAmount = Math.floor(dimDiff * (isCenterImage ? 0.5 : 1));
+          var paddingAmount = Math.round(dimDiff * (isCenterImage ? 0.5 : 1));
           var paddingAxis = height > width ? 2 : 1;
-          var getPaddingTensorShape = function (isRoundUp) {
-              if (isRoundUp === void 0) { isRoundUp = false; }
+          var createPaddingTensor = function (paddingAmount) {
               var paddingTensorShape = imgTensor.shape.slice();
-              paddingTensorShape[paddingAxis] = paddingAmount + (isRoundUp ? 1 : 0);
-              return paddingTensorShape;
+              paddingTensorShape[paddingAxis] = paddingAmount;
+              return fill(paddingTensorShape, 0);
           };
-          var tensorsToStack = (isCenterImage ? [fill(getPaddingTensorShape(!isEven(dimDiff)), 0)] : [])
-              .concat([imgTensor, fill(getPaddingTensorShape(), 0)]);
+          var paddingTensorAppend = createPaddingTensor(paddingAmount);
+          var remainingPaddingAmount = dimDiff - paddingTensorAppend.shape[paddingAxis];
+          var paddingTensorPrepend = isCenterImage && remainingPaddingAmount
+              ? createPaddingTensor(remainingPaddingAmount)
+              : null;
+          var tensorsToStack = [
+              paddingTensorPrepend,
+              imgTensor,
+              paddingTensorAppend
+          ]
+              .filter(function (t) { return t !== null; });
           return concat(tensorsToStack, paddingAxis);
       });
   }

src/padToSquare.ts

@@ -21,17 +21,28 @@ export function padToSquare(
     }
 
     const dimDiff = Math.abs(height - width)
-    const paddingAmount = Math.floor(dimDiff * (isCenterImage ? 0.5 : 1))
+    const paddingAmount = Math.round(dimDiff * (isCenterImage ? 0.5 : 1))
     const paddingAxis = height > width ? 2 : 1
 
-    const getPaddingTensorShape = (isRoundUp: boolean = false): number[] => {
+    const createPaddingTensor = (paddingAmount: number): tf.Tensor => {
       const paddingTensorShape = imgTensor.shape.slice()
-      paddingTensorShape[paddingAxis] = paddingAmount + (isRoundUp ? 1 : 0)
-      return paddingTensorShape
+      paddingTensorShape[paddingAxis] = paddingAmount
+      return tf.fill(paddingTensorShape, 0)
     }
 
-    const tensorsToStack = (isCenterImage ? [tf.fill(getPaddingTensorShape(!isEven(dimDiff)), 0)] : [])
-      .concat([imgTensor,  tf.fill(getPaddingTensorShape(), 0)]) as tf.Tensor4D[]
+    const paddingTensorAppend = createPaddingTensor(paddingAmount)
+    const remainingPaddingAmount = dimDiff - paddingTensorAppend.shape[paddingAxis]
+
+    const paddingTensorPrepend = isCenterImage && remainingPaddingAmount
+      ? createPaddingTensor(remainingPaddingAmount)
+      : null
+
+    const tensorsToStack = [
+      paddingTensorPrepend,
+      imgTensor,
+      paddingTensorAppend
+    ]
+      .filter(t => t !== null) as tf.Tensor4D[]
     return tf.concat(tensorsToStack, paddingAxis)
   })
 }
\ No newline at end of file

test/tests/padToSquare.test.ts

@@ -7,7 +7,7 @@ describe('padToSquare', () => {
 
   describe('even size', () => {
 
-    it('is padded to square', () => tf.tidy(() => {
+    it('is padded to square by 2 columns', () => tf.tidy(() => {
       const imgTensor = tf.tensor4d(Array(24).fill(1), [1, 4, 2, 3])
       const result = padToSquare(imgTensor)
 
@@ -21,7 +21,7 @@ describe('padToSquare', () => {
       expect(paddedCols[3].dataSync()).toEqual(zeros(12))
     }))
 
-    it('is padded to square and centered', () => tf.tidy(() => {
+    it('is padded to square by 2 columns and centered', () => tf.tidy(() => {
       const imgTensor = tf.tensor4d(Array(24).fill(1), [1, 4, 2, 3])
       const result = padToSquare(imgTensor, true)
 
@@ -35,11 +35,39 @@ describe('padToSquare', () => {
       expect(paddedCols[3].dataSync()).toEqual(zeros(12))
     }))
 
+    it('is padded to square by 1 column', () => tf.tidy(() => {
+      const imgTensor = tf.tensor4d(Array(36).fill(1), [1, 4, 3, 3])
+      const result = padToSquare(imgTensor)
+
+      expect(result.shape).toEqual([1, 4, 4, 3])
+
+      const paddedCols = tf.unstack(result, 2)
+      expect(paddedCols.length).toEqual(4)
+      expect(paddedCols[0].dataSync()).toEqual(ones(12))
+      expect(paddedCols[1].dataSync()).toEqual(ones(12))
+      expect(paddedCols[2].dataSync()).toEqual(ones(12))
+      expect(paddedCols[3].dataSync()).toEqual(zeros(12))
+    }))
+
+    it('is padded to square by 1 column and centered', () => tf.tidy(() => {
+      const imgTensor = tf.tensor4d(Array(36).fill(1), [1, 4, 3, 3])
+      const result = padToSquare(imgTensor, true)
+
+      expect(result.shape).toEqual([1, 4, 4, 3])
+
+      const paddedCols = tf.unstack(result, 2)
+      expect(paddedCols.length).toEqual(4)
+      expect(paddedCols[0].dataSync()).toEqual(ones(12))
+      expect(paddedCols[1].dataSync()).toEqual(ones(12))
+      expect(paddedCols[2].dataSync()).toEqual(ones(12))
+      expect(paddedCols[3].dataSync()).toEqual(zeros(12))
+    }))
+
   })
 
   describe('uneven size', () => {
 
-    it('is padded to square', () => tf.tidy(() => {
+    it('is padded to square by 3 columns', () => tf.tidy(() => {
       const imgTensor = tf.tensor4d(Array(30).fill(1), [1, 5, 2, 3])
       const result = padToSquare(imgTensor)
 
@@ -54,17 +82,46 @@ describe('padToSquare', () => {
       expect(paddedCols[4].dataSync()).toEqual(zeros(15))
     }))
 
-    it('is padded to square and centered', () => tf.tidy(() => {
+    it('is padded to square by 3 columns and centered', () => tf.tidy(() => {
       const imgTensor = tf.tensor4d(Array(30).fill(1), [1, 5, 2, 3])
       const result = padToSquare(imgTensor, true)
 
       expect(result.shape).toEqual([1, 5, 5, 3])
-      const data = result.dataSync()
 
       const paddedCols = tf.unstack(result, 2)
       expect(paddedCols.length).toEqual(5)
       expect(paddedCols[0].dataSync()).toEqual(zeros(15))
-      expect(paddedCols[1].dataSync()).toEqual(zeros(15))
+      expect(paddedCols[1].dataSync()).toEqual(ones(15))
+      expect(paddedCols[2].dataSync()).toEqual(ones(15))
+      expect(paddedCols[3].dataSync()).toEqual(zeros(15))
+      expect(paddedCols[4].dataSync()).toEqual(zeros(15))
+    }))
+
+    it('is padded to square by 1 column', () => tf.tidy(() => {
+      const imgTensor = tf.tensor4d(Array(60).fill(1), [1, 5, 4, 3])
+      const result = padToSquare(imgTensor)
+
+      expect(result.shape).toEqual([1, 5, 5, 3])
+
+      const paddedCols = tf.unstack(result, 2)
+      expect(paddedCols.length).toEqual(5)
+      expect(paddedCols[0].dataSync()).toEqual(ones(15))
+      expect(paddedCols[1].dataSync()).toEqual(ones(15))
+      expect(paddedCols[2].dataSync()).toEqual(ones(15))
+      expect(paddedCols[3].dataSync()).toEqual(ones(15))
+      expect(paddedCols[4].dataSync()).toEqual(zeros(15))
+    }))
+
+    it('is padded to square by 1 column and centered', () => tf.tidy(() => {
+      const imgTensor = tf.tensor4d(Array(60).fill(1), [1, 5, 4, 3])
+      const result = padToSquare(imgTensor, true)
+
+      expect(result.shape).toEqual([1, 5, 5, 3])
+
+      const paddedCols = tf.unstack(result, 2)
+      expect(paddedCols.length).toEqual(5)
+      expect(paddedCols[0].dataSync()).toEqual(ones(15))
+      expect(paddedCols[1].dataSync()).toEqual(ones(15))
       expect(paddedCols[2].dataSync()).toEqual(ones(15))
       expect(paddedCols[3].dataSync()).toEqual(ones(15))
       expect(paddedCols[4].dataSync()).toEqual(zeros(15))