[MD] Debug Mode - Mappable Op Row Content Order Fix and UTs

mindspore/ccsrc/minddata/dataset/engine/datasetops/source/mappable_leaf_op.cc

@@ -183,15 +183,16 @@ Status MappableLeafOp::GetNextRowPullMode(TensorRow *const row) {
     RETURN_IF_NOT_OK(sampler_->GetNextSample(&sample_row));
     CHECK_FAIL_RETURN_UNEXPECTED(sample_row.size() > 0, "GetNextRowPullMode: Expect at least one sample in sampler.");
     sample_ids_ = sample_row[0];
+    MS_LOG(DEBUG) << "Set sample_ids_=" << (*sample_ids_);
   }
   if (curr_row_ + 1 > sample_ids_->Size()) {
     *row = TensorRow(TensorRow::kFlagEOE);
     RETURN_IF_NOT_OK(ResetAndUpdateRepeat());
-    RETURN_IF_NOT_OK(sampler_->GetNextSample(&sample_row));
     return Status::OK();
   }
   int64_t key;
   RETURN_IF_NOT_OK(sample_ids_->GetItemAt(&key, {curr_row_}));
+  MS_LOG(DEBUG) << "Got key=" << key << " with curr_row_=" << curr_row_;
   RETURN_IF_NOT_OK(LoadTensorRowPullMode(key, row));
   curr_row_++;
   return Status::OK();
@@ -200,6 +201,12 @@ Status MappableLeafOp::GetNextRowPullMode(TensorRow *const row) {
 Status MappableLeafOp::ResetAndUpdateRepeat() {
   if (!IsLastIteration()) {
     RETURN_IF_NOT_OK(Reset());
+    TensorRow sample_row;
+    RETURN_IF_NOT_OK(sampler_->GetNextSample(&sample_row));
+    CHECK_FAIL_RETURN_UNEXPECTED(sample_row.size() > 0, "GetNextRowPullMode: Expect at least one sample in sampler.");
+    // Get sample_ids
+    sample_ids_ = sample_row[0];
+    MS_LOG(DEBUG) << "Set sample_ids_=" << (*sample_ids_);
     UpdateRepeatAndEpochCounter();
   } else {
     eof_handled_ = true;

tests/ut/cpp/dataset/c_api_datasets_test.cc

@@ -48,7 +48,7 @@ TEST_F(MindDataTestPipeline, TestCelebADataset) {
   ASSERT_OK(iter->GetNextRow(&row));
 
   // Check if CelebA() read correct images/attr
-  std::string expect_file[] = {"1.JPEG", "2.jpg"};
+  std::string expect_file[] = {"1_apple.JPEG", "2_banana.jpg"};
   std::vector<std::vector<uint32_t>> expect_attr_vector = {
      {0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 1, 1,
       0, 1, 0, 0, 1, 0, 0, 1, 0, 0, 0, 1, 1, 0, 1, 0, 1, 0, 0, 1},

tests/ut/data/dataset/testCelebAData/list_attr_celeba.txt

@@ -1,6 +1,6 @@
 4
 5_o_Clock_Shadow Arched_Eyebrows Attractive Bags_Under_Eyes Bald Bangs Big_Lips Big_Nose Black_Hair Blond_Hair Blurry Brown_Hair Bushy_Eyebrows Chubby Double_Chin Eyeglasses Goatee Gray_Hair Heavy_Makeup High_Cheekbones Male Mouth_Slightly_Open Mustache Narrow_Eyes No_Beard Oval_Face Pale_Skin Pointy_Nose Receding_Hairline Rosy_Cheeks Sideburns Smiling Straight_Hair Wavy_Hair Wearing_Earrings Wearing_Hat Wearing_Lipstick Wearing_Necklace Wearing_Necktie Young
-1.JPEG -1  1  1 -1 -1 -1 -1 -1 -1 -1 -1  1 -1 -1 -1 -1 -1 -1  1  1 -1  1 -1 -1  1 -1 -1  1 -1 -1 -1  1  1 -1  1 -1  1 -1 -1  1
-2.jpg  -1 -1 -1  1 -1 -1 -1  1 -1 -1 -1  1 -1 -1 -1 -1 -1 -1 -1  1 -1  1 -1 -1  1 -1 -1 -1 -1 -1 -1  1 -1 -1 -1 -1 -1 -1 -1  1
-2.jpg  -1 -1 -1  1 -1 -1 -1  1 -1 -1 -1  1 -1 -1 -1 -1 -1 -1 -1  1 -1  1 -1 -1  1 -1 -1 -1 -1 -1 -1  1 -1 -1 -1 -1 -1 -1 -1  1
-1.JPEG -1  1  1 -1 -1 -1 -1 -1 -1 -1 -1  1 -1 -1 -1 -1 -1 -1  1  1 -1  1 -1 -1  1 -1 -1  1 -1 -1 -1  1  1 -1  1 -1  1 -1 -1  1
+1_apple.JPEG -1  1  1 -1 -1 -1 -1 -1 -1 -1 -1  1 -1 -1 -1 -1 -1 -1  1  1 -1  1 -1 -1  1 -1 -1  1 -1 -1 -1  1  1 -1  1 -1  1 -1 -1  1
+2_banana.jpg  -1 -1 -1  1 -1 -1 -1  1 -1 -1 -1  1 -1 -1 -1 -1 -1 -1 -1  1 -1  1 -1 -1  1 -1 -1 -1 -1 -1 -1  1 -1 -1 -1 -1 -1 -1 -1  1
+3_lemon.jpg  -1 -1 -1  1 -1 -1 -1  1 -1 -1 -1  1 -1 -1 -1 -1 -1 -1 -1  1 -1  1 -1 -1  1 -1 -1 -1 -1 -1 -1  1 -1 -1 -1 -1 -1 -1 -1  1
+4_avacado.JPEG -1  1  1 -1 -1 -1 -1 -1 -1 -1 -1  1 -1 -1 -1 -1 -1 -1  1  1 -1  1 -1 -1  1 -1 -1  1 -1 -1 -1  1  1 -1  1 -1  1 -1 -1  1

tests/ut/data/dataset/testCelebAData/list_eval_partition.txt

@@ -1,4 +1,4 @@
-1.JPEG 0
-2.jpeg 1
-2.jpeg 2 
-2.jpeg 0 
+1_apple.JPEG 0
+2_banana.jpg 1
+3_lemon.jpg 2
+4_avacado.JPEG 0

tests/ut/data/dataset/testImageNetData4/datasetSchema.json

@@ -0,0 +1,16 @@
+{
+  "datasetType": "IMAGENET",
+  "numRows": 7,
+  "columns": {
+    "image": {
+      "type": "uint8",
+      "rank": 1,
+      "t_impl": "cvmat"
+    },
+    "label" : {
+      "type": "uint32",
+      "rank": 0,
+      "t_impl" : "flex"
+    }
+  }
+}

tests/ut/python/dataset/test_pipeline_debug_mode_multi_epoch.py

@@ -15,19 +15,74 @@
 """
 Test multiple epoch scenarios in debug mode
 """
+import math
+import numpy as np
 import pytest
 import mindspore.dataset as ds
 from mindspore import log as logger
-from util import config_get_set_seed
+from util import config_get_set_seed, visualize_list, config_get_set_num_parallel_workers
 
 pytestmark = pytest.mark.forked
 
 
+@pytest.mark.parametrize("my_debug_mode", (False, True))
+def test_pipeline_debug_mode_multi_epoch_celaba(my_debug_mode, plot=False):
+    """
+    Feature: Pipeline debug mode.
+    Description: Test multiple epoch scenario using CelebADataset.
+    Expectation: Output is equal to the expected output
+    """
+    logger.info("test_pipeline_debug_mode_multi_epoch_celaba")
+
+    # Set configuration
+    original_seed = config_get_set_seed(99)
+    original_num_workers = config_get_set_num_parallel_workers(1)
+    if my_debug_mode:
+        debug_mode_original = ds.config.get_debug_mode()
+        ds.config.set_debug_mode(True)
+
+    # testCelebAData has 4 samples
+    num_samples = 4
+
+    data1 = ds.CelebADataset("../data/dataset/testCelebAData/", decode=True)
+
+    # Confirm dataset size
+    assert data1.get_dataset_size() == num_samples
+
+    num_epoch = 2
+    iter1 = data1.create_dict_iterator(num_epochs=num_epoch, output_numpy=True)
+    epoch_count = 0
+    sample_count = 0
+    image_list = []
+    for _ in range(num_epoch):
+        row_count = 0
+        for row_item in iter1:
+            # Note: Each row in this CelebADataset pipeline has columns "image" and "attr"
+            assert len(row_item) == 2
+            assert row_item["image"].shape == (2268, 4032, 3)
+            image = row_item["image"]
+            image_list.append(image)
+            row_count += 1
+        assert row_count == num_samples
+        epoch_count += 1
+        sample_count += row_count
+    assert epoch_count == num_epoch
+    assert sample_count == num_samples * num_epoch
+    if plot:
+        visualize_list(image_list)
+
+    # Restore configuration
+    ds.config.set_seed(original_seed)
+    ds.config.set_num_parallel_workers(original_num_workers)
+    if my_debug_mode:
+        ds.config.set_debug_mode(debug_mode_original)
+
+
 @pytest.mark.parametrize("my_debug_mode", (False, True))
 def test_pipeline_debug_mode_multi_epoch_celaba_take(my_debug_mode):
     """
     Feature: Pipeline debug mode.
-    Description: Test creating tuple iterator in CelebA dataset with take op and multi epochs.
+    Description: Test multiple epoch scenario using CelebADataset with take op.
     Expectation: Output is equal to the expected output
     """
     logger.info("test_pipeline_debug_mode_multi_epoch_celaba_take")
@@ -45,14 +100,15 @@ def test_pipeline_debug_mode_multi_epoch_celaba_take(my_debug_mode):
     data1 = data1.take(num_take)
 
     num_epoch = 2
-    # Use create_tuple_iterator
-    iter1 = data1.create_tuple_iterator(num_epochs=num_epoch)
+    iter1 = data1.create_dict_iterator(num_epochs=num_epoch, output_numpy=True)
     epoch_count = 0
     sample_count = 0
     for _ in range(num_epoch):
         row_count = 0
-        for _ in iter1:
-            # in this example, each row has columns "image" and "label"
+        for row_item in iter1:
+            # Note: Each row in this CelebADataset pipeline has columns "image" and "attr"
+            assert len(row_item) == 2
+            assert row_item["image"].shape == (2268, 4032, 3)
             row_count += 1
         assert row_count == num_take
         epoch_count += 1
@@ -100,7 +156,9 @@ def test_pipeline_debug_mode_multi_epoch_cifar10_take(my_debug_mode):
     sample_count = 0
     for _ in range(num_epoch):
         row_count = 0
-        for _ in iter1:
+        for row_item in iter1:
+            image = row_item["image"]
+            assert image.shape == (32, 32, 3)
             row_count += 1
         assert row_count == num_take
         epoch_count += 1
@@ -130,29 +188,47 @@ def test_pipeline_debug_mode_multi_epoch_cifar10_repeat_batch(my_debug_mode):
         ds.config.set_debug_mode(True)
 
     data_dir_10 = "../data/dataset/testCifar10Data"
-    num_samples = 100
+    num_samples = 40
     num_repeat = 2
-    batch_size = 32
+    batch_size = 16
 
     data1 = ds.Cifar10Dataset(data_dir_10, num_samples=num_samples)
     # Add repeat then batch
     data1 = data1.repeat(num_repeat)
     data1 = data1.batch(batch_size, True)
 
-    num_epoch = 5
-    # Use create_tuple_iterator
-    iter1 = data1.create_tuple_iterator(num_epochs=num_epoch)
+    num_epoch = 2
+    iter1 = data1.create_dict_iterator(num_epochs=num_epoch, output_numpy=True)
     epoch_count = 0
     sample_count = 0
+    label_list = []
+    label_golden = [[0, 7, 8, 4, 9, 1, 9, 8, 6, 2, 7, 0, 2, 1, 7, 0],
+                    [1, 4, 4, 7, 9, 4, 7, 8, 3, 4, 3, 9, 4, 7, 3, 9],
+                    [0, 0, 5, 6, 6, 6, 5, 4, 7, 6, 9, 0, 3, 4, 3, 7],
+                    [2, 4, 0, 3, 1, 7, 7, 9, 9, 8, 7, 0, 6, 3, 7, 9],
+                    [0, 8, 7, 8, 1, 6, 4, 2, 6, 4, 9, 5, 3, 2, 5, 9],
+                    [2, 5, 3, 6, 2, 0, 7, 8, 3, 9, 6, 2, 4, 1, 5, 4],
+                    [5, 0, 9, 8, 6, 4, 9, 2, 0, 0, 0, 2, 7, 2, 6, 4],
+                    [6, 2, 7, 7, 2, 6, 9, 2, 8, 5, 7, 6, 6, 4, 6, 2],
+                    [0, 7, 2, 9, 8, 6, 7, 6, 0, 3, 2, 2, 6, 8, 8, 2],
+                    [4, 6, 6, 9, 8, 4, 6, 1, 0, 5, 5, 9, 2, 0, 8, 7]]
     for _ in range(num_epoch):
         row_count = 0
-        for _ in iter1:
+        label_list_per_epoch = []
+        for row_item in iter1:
+            image = row_item["image"]
+            label = row_item["label"]
+            assert image.shape == (batch_size, 32, 32, 3)
+            label_list.append(label)
+            label_list_per_epoch.append(label)
             row_count += 1
+        logger.info("epoch_count is {}, label_list_per_epoch is {}".format(epoch_count, label_list_per_epoch))
         assert row_count == int(num_samples * num_repeat / batch_size)
         epoch_count += 1
         sample_count += row_count
     assert epoch_count == num_epoch
     assert sample_count == int(num_samples * num_repeat / batch_size) * num_epoch
+    np.testing.assert_array_equal(label_list, np.array(label_golden))
 
     # Restore configuration
     ds.config.set_seed(original_seed)
@@ -190,8 +266,14 @@ def test_pipeline_debug_mode_multi_epoch_cifar10_batch_repeat(my_debug_mode):
     sample_count = 0
     for _ in range(num_epoch):
         row_count = 0
-        for _ in iter1:
+        label_list_per_epoch = []
+        for row_item in iter1:
+            image = row_item["image"]
+            label = row_item["label"]
+            assert image.shape == (batch_size, 32, 32, 3)
+            label_list_per_epoch.append(label)
             row_count += 1
+        logger.info("epoch_count is {}, label_list_per_epoch is {}".format(epoch_count, label_list_per_epoch))
         assert row_count == int(num_samples * num_repeat / batch_size)
         epoch_count += 1
         sample_count += row_count
@@ -242,8 +324,14 @@ def test_pipeline_debug_mode_multi_epoch_cifar10_zip(my_debug_mode):
     sample_count = 0
     for _ in range(num_epoch):
         row_count = 0
-        for _ in iter1:
+        label_list_per_epoch = []
+        for row_item in iter1:
+            image = row_item["image"]
+            label = row_item["label"]
+            assert image.shape == (batch_size, 32, 32, 3)
+            label_list_per_epoch.append(label)
             row_count += 1
+        logger.info("epoch_count is {}, label_list_per_epoch is {}".format(epoch_count, label_list_per_epoch))
         assert row_count == int(num_samples * num_repeat / batch_size)
         epoch_count += 1
         sample_count += row_count
@@ -274,7 +362,7 @@ def test_pipeline_debug_mode_multi_epoch_cifar10_zip_batch_repeat(my_debug_mode)
     data_dir_10 = "../data/dataset/testCifar10Data"
     num_samples = 20
     batch_size = 10
-    num_repeat = 5
+    num_repeat = 3
 
     data1 = ds.Cifar10Dataset(data_dir_10, num_samples=num_samples)
 
@@ -292,15 +380,36 @@ def test_pipeline_debug_mode_multi_epoch_cifar10_zip_batch_repeat(my_debug_mode)
     iter1 = data3.create_dict_iterator(num_epochs=num_epoch, output_numpy=True)
     epoch_count = 0
     sample_count = 0
+    label_list = []
+    label_golden = [[5, 1, 3, 6, 2, 7, 5, 2, 1, 9],
+                    [3, 0, 9, 1, 1, 2, 5, 5, 6, 3],
+                    [5, 5, 8, 0, 8, 5, 4, 7, 2, 2],
+                    [2, 2, 4, 8, 1, 1, 3, 0, 5, 8],
+                    [1, 1, 0, 5, 5, 5, 8, 4, 4, 1],
+                    [8, 2, 9, 0, 8, 1, 6, 0, 1, 8],
+                    [7, 0, 6, 1, 6, 2, 7, 4, 2, 3],
+                    [9, 8, 0, 2, 7, 4, 1, 9, 8, 3],
+                    [7, 0, 2, 6, 2, 0, 2, 0, 7, 0],
+                    [4, 7, 7, 7, 6, 5, 3, 4, 5, 9],
+                    [1, 9, 7, 5, 7, 7, 2, 2, 9, 2],
+                    [8, 8, 5, 1, 4, 0, 5, 5, 6, 6]]
     for _ in range(num_epoch):
         row_count = 0
-        for _ in iter1:
+        label_list_per_epoch = []
+        for row_item in iter1:
+            image = row_item["image"]
+            label = row_item["label"]
+            assert image.shape == (batch_size, 32, 32, 3)
+            label_list.append(label)
+            label_list_per_epoch.append(label)
             row_count += 1
+        logger.info("epoch_count is {}, label_list_per_epoch is {}".format(epoch_count, label_list_per_epoch))
         assert row_count == int(num_samples * num_repeat / batch_size)
         epoch_count += 1
         sample_count += row_count
     assert epoch_count == num_epoch
     assert sample_count == int(num_samples * num_repeat / batch_size) * num_epoch
+    np.testing.assert_array_equal(label_list, np.array(label_golden))
 
     # Restore configuration
     ds.config.set_seed(original_seed)
@@ -308,53 +417,322 @@ def test_pipeline_debug_mode_multi_epoch_cifar10_zip_batch_repeat(my_debug_mode)
         ds.config.set_debug_mode(debug_mode_original)
 
 
-@pytest.mark.parametrize("my_debug_mode, my_drop_remainder, my_num_samples",
-                         [(False, False, 40), (True, False, 40), (True, True, 43)])
-def test_pipeline_debug_mode_multi_epoch_imagefolder_batch(my_debug_mode, my_drop_remainder, my_num_samples):
+@pytest.mark.parametrize("my_debug_mode", (False, True))
+def test_pipeline_debug_mode_multi_epoch_imagefolder(my_debug_mode, plot=False):
     """
     Feature: Pipeline debug mode.
-    Description: Test multiple epoch scenario using ImageFolderDataset with batch op.
+    Description: Test multiple epoch scenario using ImageFolderDataset. Plot support provided.
     Expectation: Output is equal to the expected output
     """
-    logger.info("test_pipeline_debug_mode_multi_epoch_imagefolder_batch")
+    logger.info("test_pipeline_debug_mode_multi_epoch_imagefolder")
 
     # Set configuration
-    original_seed = config_get_set_seed(799)
+    original_seed = config_get_set_seed(899)
+    original_num_workers = config_get_set_num_parallel_workers(1)
+    if my_debug_mode:
+        debug_mode_original = ds.config.get_debug_mode()
+        ds.config.set_debug_mode(True)
+
+    # Note: testImageNetData4 has 7 samples in total
+    num_samples = 7
+
+    # Use all 7 samples from the dataset
+    data1 = ds.ImageFolderDataset("../data/dataset/testImageNetData4/train",
+                                  decode=True)
+    # Confirm dataset size
+    assert data1.get_dataset_size() == num_samples
+
+    num_epoch = 4
+    iter1 = data1.create_dict_iterator(num_epochs=num_epoch, output_numpy=True)
+    epoch_count = 0
+    sample_count = 0
+    image_list = []
+    label_list = []
+    label_golden = [4, 3, 5, 0, 1, 2, 6] + [3, 2, 5, 1, 0, 6, 4] + [6, 0, 1, 2, 5, 4, 3] + [3, 4, 5, 1, 0, 6, 2]
+    for _ in range(num_epoch):
+        row_count = 0
+        label_list_per_epoch = []
+        for row_item in iter1:
+            image = row_item["image"]
+            label = row_item["label"]
+            assert image.shape == (384, 682, 3)
+            image_list.append(image)
+            label_list.append(label)
+            label_list_per_epoch.append(label)
+            row_count += 1
+        logger.info("epoch_count is {}, label_list_per_epoch is {}".format(epoch_count, label_list_per_epoch))
+        assert row_count == num_samples
+        epoch_count += 1
+        sample_count += row_count
+    assert epoch_count == num_epoch
+    assert sample_count == num_samples * num_epoch
+    assert label_list == label_golden
+    if plot:
+        visualize_list(image_list)
+
+    # Restore configuration
+    ds.config.set_seed(original_seed)
+    ds.config.set_num_parallel_workers(original_num_workers)
+    if my_debug_mode:
+        ds.config.set_debug_mode(debug_mode_original)
+
+
+@pytest.mark.parametrize("my_debug_mode, my_shuffle", [(False, True), (True, None), (True, True), (True, False)])
+def test_pipeline_debug_mode_multi_epoch_imagefolder_shuffle(my_debug_mode, my_shuffle, plot=False):
+    """
+    Feature: Pipeline debug mode.
+    Description: Test multiple epoch scenario using ImageFolderDataset with shuffle parameter. Plot support provided.
+    Expectation: Output is equal to the expected output
+    """
+    logger.info("test_pipeline_debug_mode_multi_epoch_imagefolder_shuffle")
+
+    # Set configuration
+    original_seed = config_get_set_seed(899)
+    original_num_workers = config_get_set_num_parallel_workers(1)
+    if my_debug_mode:
+        debug_mode_original = ds.config.get_debug_mode()
+        ds.config.set_debug_mode(True)
+
+    num_samples = 5
+    data1 = ds.ImageFolderDataset("../data/dataset/testImageNetData4/train",
+                                  shuffle=my_shuffle,
+                                  num_samples=num_samples,
+                                  decode=True)
+
+    num_epoch = 2
+    iter1 = data1.create_dict_iterator(num_epochs=num_epoch, output_numpy=True)
+    epoch_count = 0
+    sample_count = 0
+    image_list = []
+    label_list = []
+    if my_shuffle is False:
+        # Sequential order is used
+        label_golden = list(range(0, 5)) * num_epoch
+    else:
+        # Random order is used, according to the seed value
+        label_golden = [2, 1, 3, 0, 4] + [3, 6, 2, 0, 3]
+    for _ in range(num_epoch):
+        row_count = 0
+        label_list_per_epoch = []
+        for row_item in iter1:
+            image = row_item["image"]
+            label = row_item["label"]
+            assert image.shape == (384, 682, 3)
+            image_list.append(image)
+            label_list.append(label)
+            label_list_per_epoch.append(label)
+            row_count += 1
+        logger.info("epoch_count is {}, label_list_per_epoch is {}".format(epoch_count, label_list_per_epoch))
+        assert row_count == num_samples
+        epoch_count += 1
+        sample_count += row_count
+    assert epoch_count == num_epoch
+    assert sample_count == num_samples * num_epoch
+    assert label_list == label_golden
+    if plot:
+        visualize_list(image_list)
+
+    # Restore configuration
+    ds.config.set_seed(original_seed)
+    ds.config.set_num_parallel_workers(original_num_workers)
+    if my_debug_mode:
+        ds.config.set_debug_mode(debug_mode_original)
+
+
+@pytest.mark.parametrize("my_debug_mode", (False, True))
+def test_pipeline_debug_mode_multi_epoch_imagefolder_repeat(my_debug_mode, plot=False):
+    """
+    Feature: Pipeline debug mode.
+    Description: Test multiple epoch scenario using ImageFolderDataset with repeat op. Plot support provided.
+    Expectation: Output is equal to the expected output
+    """
+    logger.info("test_pipeline_debug_mode_multi_epoch_imagefolder_repeat")
+
+    # Set configuration
+    original_seed = config_get_set_seed(899)
+    original_num_workers = config_get_set_num_parallel_workers(1)
+    if my_debug_mode:
+        debug_mode_original = ds.config.get_debug_mode()
+        ds.config.set_debug_mode(True)
+
+    num_samples = 5
+    num_repeat = 3
+    data1 = ds.ImageFolderDataset("../data/dataset/testImageNetData4/train",
+                                  shuffle=True,
+                                  num_samples=num_samples,
+                                  decode=True)
+    data1 = data1.repeat(num_repeat)
+
+    num_epoch = 2
+    iter1 = data1.create_dict_iterator(num_epochs=num_epoch, output_numpy=True)
+    epoch_count = 0
+    sample_count = 0
+    image_list = []
+    label_list = []
+    # Random order is used, according to the seed value
+    label_golden = [2, 1, 3, 0, 4] + [3, 6, 2, 0, 3] + \
+                   [5, 4, 0, 1, 0] + [0, 0, 1, 3, 5] + \
+                   [4, 5, 5, 2, 0] + [1, 2, 4, 4, 5]
+    for _ in range(num_epoch):
+        row_count = 0
+        label_list_per_epoch = []
+        for row_item in iter1:
+            image = row_item["image"]
+            label = row_item["label"]
+            assert image.shape == (384, 682, 3)
+            image_list.append(image)
+            label_list.append(label)
+            label_list_per_epoch.append(label)
+            row_count += 1
+        logger.info("epoch_count is {}, label_list_per_epoch is {}".format(epoch_count, label_list_per_epoch))
+        assert row_count == num_samples * num_repeat
+        epoch_count += 1
+        sample_count += row_count
+    assert epoch_count == num_epoch
+    assert sample_count == num_samples * num_repeat * num_epoch
+    assert label_list == label_golden
+
+    if plot:
+        visualize_list(image_list)
+
+    # Restore configuration
+    ds.config.set_seed(original_seed)
+    ds.config.set_num_parallel_workers(original_num_workers)
+    if my_debug_mode:
+        ds.config.set_debug_mode(debug_mode_original)
+
+
+@pytest.mark.parametrize("my_debug_mode, my_drop, my_num_samples",
+                         [(False, False, 6), (True, False, 6), (True, True, 7)])
+def test_pipeline_debug_mode_multi_ep_im_batch_no_remainder(my_debug_mode, my_drop, my_num_samples, plot=False):
+    """
+    Feature: Pipeline debug mode.
+    Description: Test multiple epoch scenario using ImageFolderDataset with batch op and no remainder.
+    Expectation: Output is equal to the expected output
+    """
+    logger.info("test_pipeline_debug_mode_multi_ep_im_batch_no_remainder")
+
+    # Set configuration
+    original_seed = config_get_set_seed(899)
+    original_num_workers = config_get_set_num_parallel_workers(1)
     if my_debug_mode:
         debug_mode_original = ds.config.get_debug_mode()
         ds.config.set_debug_mode(True)
 
     num_samples = my_num_samples
-    batch_size = 5
+    batch_size = 2
 
-    data1 = ds.ImageFolderDataset("../data/dataset/testPK/data", num_samples=num_samples)
-    data1 = data1.batch(batch_size, drop_remainder=my_drop_remainder)
+    data1 = ds.ImageFolderDataset("../data/dataset/testImageNetData4/train",
+                                  num_samples=num_samples,
+                                  decode=True)
+    data1 = data1.batch(batch_size, drop_remainder=my_drop)
 
     num_epoch = 3
     iter1 = data1.create_dict_iterator(num_epochs=num_epoch, output_numpy=True)
     epoch_count = 0
     sample_count = 0
+    image_list = []
+    label_list = []
+    label_golden = [[2, 1], [3, 0], [4, 6]] + [[3, 6], [2, 0], [3, 5]] + [[5, 4], [0, 1], [0, 6]]
     for _ in range(num_epoch):
         row_count = 0
-        for _ in iter1:
+        label_list_per_epoch = []
+        for row_item in iter1:
+            image = row_item["image"]
+            label = row_item["label"]
+            assert image.shape == (2, 384, 682, 3)
+            image_list.append(image[0])
+            label_list.append(label)
+            label_list_per_epoch.append(label)
             row_count += 1
+        logger.info("epoch_count is {}, label_list_per_epoch is {}".format(epoch_count, label_list_per_epoch))
         assert row_count == int(num_samples / batch_size)
         epoch_count += 1
         sample_count += row_count
+
     assert epoch_count == num_epoch
     assert sample_count == int(num_samples / batch_size) * num_epoch
+    np.testing.assert_array_equal(label_list, np.array(label_golden))
+    if plot:
+        visualize_list(image_list)
 
     # Restore configuration
     ds.config.set_seed(original_seed)
+    ds.config.set_num_parallel_workers(original_num_workers)
+    if my_debug_mode:
+        ds.config.set_debug_mode(debug_mode_original)
+
+
+@pytest.mark.parametrize("my_debug_mode, my_drop, my_num_samples",
+                         [(False, False, 7), (True, False, 7)])
+def test_pipeline_debug_mode_multi_ep_im_batch_with_remainders(my_debug_mode, my_drop, my_num_samples, plot=False):
+    """
+    Feature: Pipeline debug mode.
+    Description: Test multiple epoch scenario using ImageFolderDataset with batch op and remainder.
+    Expectation: Output is equal to the expected output
+    """
+    logger.info("test_pipeline_debug_mode_multi_ep_im_batch_with_remainders")
+
+    # Set configuration
+    original_seed = config_get_set_seed(899)
+    original_num_workers = config_get_set_num_parallel_workers(1)
+    if my_debug_mode:
+        debug_mode_original = ds.config.get_debug_mode()
+        ds.config.set_debug_mode(True)
+
+    num_samples = my_num_samples
+    batch_size = 2
+
+    data1 = ds.ImageFolderDataset("../data/dataset/testImageNetData4/train",
+                                  num_samples=num_samples,
+                                  decode=True)
+    data1 = data1.batch(batch_size, drop_remainder=my_drop)
+
+    num_epoch = 3
+    iter1 = data1.create_dict_iterator(num_epochs=num_epoch, output_numpy=True)
+    epoch_count = 0
+    sample_count = 0
+    image_list = []
+    label_list = []
+    label_golden = [[2, 1], [3, 0], [4, 6], [5]] + [[3, 6], [2, 0], [3, 5], [1]] + [[5, 4], [0, 1], [0, 6], [6]]
+    for _ in range(num_epoch):
+        row_count = 0
+        label_list_per_epoch = []
+        for row_item in iter1:
+            image = row_item["image"]
+            label = row_item["label"]
+            assert image.shape == (2, 384, 682, 3) or (1, 384, 682, 3)
+            image_list.append(image[0])
+            label_list.append(list(label))
+            label_list_per_epoch.append(list(label))
+            row_count += 1
+        logger.info("epoch_count is {}, label_list_per_epoch is {}".format(epoch_count, label_list_per_epoch))
+        assert row_count == math.ceil(num_samples / batch_size)
+        epoch_count += 1
+        sample_count += row_count
+    assert epoch_count == num_epoch
+    assert sample_count == math.ceil(num_samples / batch_size) * num_epoch
+    assert label_list == label_golden
+    if plot:
+        visualize_list(image_list)
+
+    # Restore configuration
+    ds.config.set_seed(original_seed)
+    ds.config.set_num_parallel_workers(original_num_workers)
     if my_debug_mode:
         ds.config.set_debug_mode(debug_mode_original)
 
 
 if __name__ == '__main__':
+    test_pipeline_debug_mode_multi_epoch_celaba(True, plot=True)
     test_pipeline_debug_mode_multi_epoch_celaba_take(True)
     test_pipeline_debug_mode_multi_epoch_cifar10_take(True)
     test_pipeline_debug_mode_multi_epoch_cifar10_repeat_batch(True)
     test_pipeline_debug_mode_multi_epoch_cifar10_batch_repeat(True)
     test_pipeline_debug_mode_multi_epoch_cifar10_zip(True)
     test_pipeline_debug_mode_multi_epoch_cifar10_zip_batch_repeat(True)
-    test_pipeline_debug_mode_multi_epoch_imagefolder_batch(True, False, 10)
+    test_pipeline_debug_mode_multi_epoch_imagefolder(True, plot=True)
+    test_pipeline_debug_mode_multi_epoch_imagefolder_shuffle(True, True, plot=True)
+    test_pipeline_debug_mode_multi_epoch_imagefolder_repeat(True, plot=True)
+    test_pipeline_debug_mode_multi_ep_im_batch_no_remainder(True, True, 7, plot=True)
+    test_pipeline_debug_mode_multi_ep_im_batch_with_remainder(True, False, 7, plot=True)