add np.frombuffer to support return bytes

mindspore/dataset/engine/datasets.py

@@ -3502,11 +3502,11 @@ def _iter_fn(dataset, num_samples):
             except StopIteration:
                 return
             # convert output tensors to ndarrays
-            yield tuple([np.array(x, copy=False) for x in val])
+            yield _convert_row(val)
     else:
         for val in dataset:
             # convert output tensors to ndarrays
-            yield tuple([np.array(x, copy=False) for x in val])
+            yield _convert_row(val)
 
 
 def _generator_fn(generator, num_samples):
@@ -3539,7 +3539,7 @@ def _cpp_sampler_fn(sample_ids, dataset):
     for i in sample_ids:
         val = dataset[i]
         # convert output tensors to ndarrays
-        yield tuple([np.array(x, copy=False) for x in val])
+        yield _convert_row(val)
 
 
 def _cpp_sampler_fn_mp(sample_ids, sample_fn):
@@ -3606,6 +3606,17 @@ def _watch_dog(pids, eof):
             os.kill(os.getpid(), signal.SIGTERM)
 
 
+def _convert_row(row):
+    value = []
+    # convert each column in row into numpy array
+    for x in row:
+        if isinstance(x, bytes):
+            value.append(np.frombuffer(x, np.uint8))
+        else:
+            value.append(np.array(x, copy=False))
+    return tuple(value)
+
+
 class SamplerFn:
     """
     Multiprocessing or multithread generator function wrapper master process.
@@ -3704,7 +3715,7 @@ class SamplerFn:
                 return
             if idx_cursor < len(indices):
                 idx_cursor = _fill_worker_indices(self.workers, indices, idx_cursor)
-            yield tuple([np.array(x, copy=False) for x in result])
+            yield _convert_row(result)
 
     def _stop_subprocess(self):
         # Only the main process can call join

tests/ut/python/dataset/test_decode.py

@@ -16,9 +16,11 @@
 Testing Decode op in DE
 """
 import cv2
+import numpy as np
 
 import mindspore.dataset as ds
 import mindspore.dataset.vision.c_transforms as vision
+import mindspore.dataset.vision.py_transforms as py_vision
 from mindspore import log as logger
 from util import diff_mse
 
@@ -76,6 +78,41 @@ def test_decode_op_tf_file_dataset():
         assert mse == 0
 
 
+class ImageDataset:
+    def __init__(self, data_path, data_type="numpy"):
+        self.data = [data_path]
+        self.label = np.random.sample((1, 1))
+        self.data_type = data_type
+
+    def __getitem__(self, index):
+        # use file open and read method
+        f = open(self.data[index], 'rb')
+        img_bytes = f.read()
+        f.close()
+        if self.data_type == "numpy":
+            img_bytes = np.frombuffer(img_bytes, dtype=np.uint8)
+
+        # return bytes directly
+        return (img_bytes, self.label[index])
+
+    def __len__(self):
+        return len(self.data)
+
+
+def test_read_image_decode_op():
+    data_path = "../data/dataset/testPK/data/class1/0.jpg"
+    dataset1 = ds.GeneratorDataset(ImageDataset(data_path, data_type="numpy"), ["data", "label"])
+    dataset2 = ds.GeneratorDataset(ImageDataset(data_path, data_type="bytes"), ["data", "label"])
+    decode_op = py_vision.Decode()
+    to_tensor = py_vision.ToTensor(output_type=np.int32)
+    dataset1 = dataset1.map(operations=[decode_op, to_tensor], input_columns=["data"])
+    dataset2 = dataset2.map(operations=[decode_op, to_tensor], input_columns=["data"])
+
+    for item1, item2 in zip(dataset1, dataset2):
+        assert np.count_nonzero(item1[0].asnumpy() - item2[0].asnumpy()) == 0
+
+
 if __name__ == "__main__":
     test_decode_op()
     test_decode_op_tf_file_dataset()
+    test_read_image_decode_op()