Fer2013.py issue

test/test_datasets.py

@@ -2462,6 +2462,30 @@ def inject_fake_data(self, tmpdir, config):
 
                 writer.writerow(row)
 
+        with open(os.path.join(base_folder, "fer2013.csv"), "w", newline="") as file:
+            writer = csv.DictWriter(
+                file,
+                fieldnames=("emotion", "pixels", "Usage"),
+                quoting=csv.QUOTE_NONNUMERIC,
+                quotechar='"',
+            )
+            writer.writeheader()
+            for i in range(num_samples + 28709):
+                row = dict(
+                    pixels=" ".join(
+                        str(pixel) for pixel in datasets_utils.create_image_or_video_tensor((48, 48)).view(-1).tolist()
+                    )
+                )
+                row["emotion"] = str(int(torch.randint(0, 7, ())))
+
+                usage_values = ["PrivateTest", "PublicTest"]
+                if i < 28709:
+                    row["Usage"] = "Training"
+                else:
+                    row["Usage"] = random.choice(usage_values)
+
+                writer.writerow(row)
+
         return num_samples
 
 

torchvision/datasets/fer2013.py

@@ -1,7 +1,11 @@
 import csv
 import pathlib
+import warnings
+from itertools import islice
 from typing import Any, Callable, Optional, Tuple, Union
 
+warnings.simplefilter("always", Warning)
+
 import torch
 from PIL import Image
 
@@ -27,6 +31,8 @@ class FER2013(VisionDataset):
         "test": ("test.csv", "b02c2298636a634e8c2faabbf3ea9a23"),
     }
 
+    _RESOURCES_EXTRA = {"fer2013": ("fer2013.csv", "f8428a1edbd21e88f42c73edd2a14f95")}
+
     def __init__(
         self,
         root: Union[str, pathlib.Path],
@@ -56,11 +62,49 @@ def __init__(
                 for row in csv.DictReader(file)
             ]
 
+        self.get_test_label = False
+        self.err_fer2013_warning = (
+            "The integrity of fer2013.csv has been compromised, "
+            "thus the test labels will not be provided. "
+            "You can download it from "
+            "https://www.kaggle.com/c/challenges-in-representation-learning-facial-expression-recognition-challenge"
+        )
+        if self._split == "test":
+            self.get_test_label = True
+
+        if self.get_test_label:
+            try:
+                file_name_entire, md5_entire = self._RESOURCES_EXTRA["fer2013"]
+                entire_data_file = base_folder / file_name_entire
+                if not check_integrity(str(entire_data_file), md5=md5_entire):
+                    self.get_test_label = False
+                    warnings.warn(self.err_fer2013_warning, Warning)
+            except KeyError:
+                self.get_test_label = False
+                warnings.warn(self.err_fer2013_warning, Warning)
+
+        if self.get_test_label:
+            with open(entire_data_file, "r", newline="") as file:
+                reader = csv.DictReader(file)
+                test_rows = islice(reader, 28709, None)
+                self._samples_alt = [
+                    (
+                        torch.tensor([int(idx) for idx in row["pixels"].split()], dtype=torch.uint8).reshape(48, 48),
+                        int(row["emotion"]) if "emotion" in row else None,
+                    )
+                    for row in test_rows
+                ]
+
     def __len__(self) -> int:
         return len(self._samples)
 
     def __getitem__(self, idx: int) -> Tuple[Any, Any]:
         image_tensor, target = self._samples[idx]
+
+        if self.get_test_label:
+            _, target_alt = self._samples_alt[idx]
+            target = target_alt
+
         image = Image.fromarray(image_tensor.numpy())
 
         if self.transform is not None: