torch batch boundary CE loss

Summary:
A torch version which is batched allows us to avoid CPU <--> GPU copy which
gets us ~200ms per iteration saving. This new version of generating boundary
weight mask produces identical masks.

Reviewed By: wat3rBro

Differential Revision: D30176412

fbshipit-source-id: 877f4c6337e7870d3bafd8eb9157ac166ddd588a

tests/modeling/test_trimap.py

-#!/usr/bin/env python3
-# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved
-
-
-import numpy as np
-import unittest
-
-from d2go.modeling.trimap import generate_boundary_weight_mask
-
-
-class TestTrimap(unittest.TestCase):
-
-    def test_generate_boundary_weight_mask(self):
-        # smallest size mask to not give dilation too small warning
-        mask_gt = np.zeros((115, 252))
-        mask_gt[54:60, 123:128] = 1
-
-        mask_dt = np.zeros((115, 252))
-        mask_dt[55:61, 124:129] = 1
-
-        weights = generate_boundary_weight_mask(mask_gt, mask_dt, non_bd_weight=0.1)
-        unique_values = np.unique(weights)
-        self.assertEqual(unique_values.size, 2)
-        self.assertEqual(10 * unique_values[0], unique_values[1])
-
-    def test_generate_boundary_weight_mask_from_negative_img_mask(self):
-        # smallest size mask to not give dilation too small warning
-        mask_gt = np.zeros((115, 252))
-        mask_dt = np.zeros((115, 252))
-
-        try:
-            generate_boundary_weight_mask(mask_gt, mask_dt, non_bd_weight=0.0)
-        except ZeroDivisionError:
-            self.fail(
-                "generate_boundary_weight_mask raised ZeroDivisionError unexpectedly")