[Fix] Fix the bug that when all pixels in an image is ignored, the ac… (

open-mmlab#1336)

* [Fix] Fix the bug that when all pixels in an image is ignored, the accuracy calculation raises ZeroDivisionError

* use eps

* all close

* add ignore test

* add eps

mmseg/models/losses/accuracy.py

@@ -1,4 +1,5 @@
 # Copyright (c) OpenMMLab. All rights reserved.
+import torch
 import torch.nn as nn
 
 
@@ -46,10 +47,13 @@ def accuracy(pred, target, topk=1, thresh=None, ignore_index=None):
         correct = correct & (pred_value > thresh).t()
     correct = correct[:, target != ignore_index]
     res = []
+    eps = torch.finfo(torch.float32).eps
     for k in topk:
-        correct_k = correct[:k].reshape(-1).float().sum(0, keepdim=True)
-        res.append(
-            correct_k.mul_(100.0 / target[target != ignore_index].numel()))
+        # Avoid causing ZeroDivisionError when all pixels
+        # of an image are ignored
+        correct_k = correct[:k].reshape(-1).float().sum(0, keepdim=True) + eps
+        total_num = target[target != ignore_index].numel() + eps
+        res.append(correct_k.mul_(100.0 / total_num))
     return res[0] if return_single else res
 
 

tests/test_models/test_losses/test_utils.py

@@ -56,50 +56,56 @@ def test_accuracy():
     true_label = torch.Tensor([2, 3, 0, 1, 2]).long()
     accuracy = Accuracy(topk=1, ignore_index=None)
     acc = accuracy(pred, true_label)
-    assert acc.item() == 100
+    assert torch.allclose(acc, torch.tensor(100.0))
 
     # test for ignore_index with a wrong prediction of that index
     true_label = torch.Tensor([2, 3, 1, 1, 2]).long()
     accuracy = Accuracy(topk=1, ignore_index=1)
     acc = accuracy(pred, true_label)
-    assert acc.item() == 100
+    assert torch.allclose(acc, torch.tensor(100.0))
 
     # test for ignore_index 1 with a wrong prediction of other index
     true_label = torch.Tensor([2, 0, 0, 1, 2]).long()
     accuracy = Accuracy(topk=1, ignore_index=1)
     acc = accuracy(pred, true_label)
-    assert acc.item() == 75
+    assert torch.allclose(acc, torch.tensor(75.0))
 
     # test for ignore_index 4 with a wrong prediction of other index
     true_label = torch.Tensor([2, 0, 0, 1, 2]).long()
     accuracy = Accuracy(topk=1, ignore_index=4)
     acc = accuracy(pred, true_label)
-    assert acc.item() == 80
+    assert torch.allclose(acc, torch.tensor(80.0))
+
+    # test for ignoring all the pixels
+    true_label = torch.Tensor([2, 2, 2, 2, 2]).long()
+    accuracy = Accuracy(topk=1, ignore_index=2)
+    acc = accuracy(pred, true_label)
+    assert torch.allclose(acc, torch.tensor(100.0))
 
     # test for top1
     true_label = torch.Tensor([2, 3, 0, 1, 2]).long()
     accuracy = Accuracy(topk=1)
     acc = accuracy(pred, true_label)
-    assert acc.item() == 100
+    assert torch.allclose(acc, torch.tensor(100.0))
 
     # test for top1 with score thresh=0.8
     true_label = torch.Tensor([2, 3, 0, 1, 2]).long()
     accuracy = Accuracy(topk=1, thresh=0.8)
     acc = accuracy(pred, true_label)
-    assert acc.item() == 40
+    assert torch.allclose(acc, torch.tensor(40.0))
 
     # test for top2
     accuracy = Accuracy(topk=2)
     label = torch.Tensor([3, 2, 0, 0, 2]).long()
     acc = accuracy(pred, label)
-    assert acc.item() == 100
+    assert torch.allclose(acc, torch.tensor(100.0))
 
     # test for both top1 and top2
     accuracy = Accuracy(topk=(1, 2))
     true_label = torch.Tensor([2, 3, 0, 1, 2]).long()
     acc = accuracy(pred, true_label)
     for a in acc:
-        assert a.item() == 100
+        assert torch.allclose(a, torch.tensor(100.0))
 
     # topk is larger than pred class number
     with pytest.raises(AssertionError):