Fix CUB200 Download (#406)

* Replace GDrive link by Zenodo.

* Update CHANGELOG.

* Fix pypi dependency in CI.

Author: seba-1511

.github/workflows/python_unittest.yaml

@@ -56,6 +56,7 @@ jobs:
                   python3 --version
                   python3 -m pip install -U pip
                   pip3 install cython pybind11
+                  pip3 install scipy==1.10.1
                   pip3 install torch==${{ matrix.pytorch }}
                   pip3 install torchvision==${{ matrix.torchvision }}
                   pip3 install chardet==3.0.4  # can be remove when fix in: https://github.com/aio-libs/aiohttp/issues/5366

CHANGELOG.md

@@ -28,7 +28,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 * MAML Toy example. (@[Theo Morales](https://github.com/DubiousCactus))
 * Example for `detach_module`. ([Nimish Sanghi](https://github.com/nsanghi))
 * Loading duplicate FGVC Aircraft images.
-* Move vision datasets to Zenodo. (mini-ImageNet, tiered-ImageNet, FC100, CIFAR-FS)
+* Move vision datasets to Zenodo. (mini-ImageNet, tiered-ImageNet, FC100, CIFAR-FS, CUB200)
 * mini-ImageNet targets are now ints (not np.float64).
 * Swap family for variants in FGVCAircraft, as in MetaDataset.
 

learn2learn/vision/datasets/cu_birds200.py

@@ -5,11 +5,15 @@
 import torch
 
 from PIL import Image
-from learn2learn.data.utils import download_file_from_google_drive
+from learn2learn.data.utils import (
+    download_file_from_google_drive,
+    download_file,
+)
 
 DATA_DIR = 'cubirds200'
 DATA_FILENAME = 'CUB_200_2011.tgz'
 ARCHIVE_ID = '1hbzc_P1FuxMkcabkgn9ZKinBwW683j45'
+ZENODO_URL = 'https://zenodo.org/record/8000562/files/CUB_200_2011.tgz'
 
 SPLITS = {
     'train': [
@@ -353,11 +357,18 @@ def download(self):
         os.makedirs(data_path, exist_ok=True)
         tar_path = os.path.join(data_path, DATA_FILENAME)
         print('Downloading CUBirds200 dataset. (1.1Gb)')
-        download_file_from_google_drive(ARCHIVE_ID, tar_path)
-        tar_file = tarfile.open(tar_path)
-        tar_file.extractall(data_path)
-        tar_file.close()
-        os.remove(tar_path)
+        try:
+            download_file(ZENODO_URL, tar_path)
+            tar_file = tarfile.open(tar_path)
+            tar_file.extractall(data_path)
+            tar_file.close()
+            os.remove(tar_path)
+        except Exception:
+            download_file_from_google_drive(ARCHIVE_ID, tar_path)
+            tar_file = tarfile.open(tar_path)
+            tar_file.extractall(data_path)
+            tar_file.close()
+            os.remove(tar_path)
 
     def load_data(self, mode='train'):
         classes = sum(SPLITS.values(), []) if mode == 'all' else SPLITS[mode]