Merge branch 'main' into kill-circleci

Author: pmeier

.github/scripts/setup-env.sh

@@ -27,9 +27,8 @@ if [[ "${OS_TYPE}" == "macos" && $(uname -m) == x86_64 ]]; then
   # The x86 macOS runners, e.g. the GitHub Actions native "macos-12" runner, has some JPEG and PNG libraries
   # installed by default that interfere with our build. We uninstall them here and use the one from conda below.
   IMAGE_LIBS=$(brew list | grep -E "jpeg|png")
-  echo "${IMAGE_LIBS}"
-  for lib in "${IMAGE_LIBS}"; do
-    brew uninstall --ignore-dependencies --force "${lib}" || true
+  for lib in $IMAGE_LIBS; do
+    brew uninstall --ignore-dependencies --force "${lib}"
   done
   echo '::endgroup::'
 fi

.github/workflows/prototype-tests-linux-gpu.yml

@@ -18,7 +18,7 @@ jobs:
           - python-version: "3.8"
             runner: linux.g5.4xlarge.nvidia.gpu
             gpu-arch-type: cuda
-            gpu-arch-version: "11.7"
+            gpu-arch-version: "11.8"
       fail-fast: false
     uses: pytorch/test-infra/.github/workflows/linux_job.yml@main
     with:

.github/workflows/tests.yml

@@ -24,7 +24,7 @@ jobs:
           - python-version: 3.8
             runner: linux.g5.4xlarge.nvidia.gpu
             gpu-arch-type: cuda
-            gpu-arch-version: "11.7"
+            gpu-arch-version: "11.8"
       fail-fast: false
     uses: pytorch/test-infra/.github/workflows/linux_job.yml@main
     with:
@@ -85,7 +85,7 @@ jobs:
           - python-version: "3.8"
             runner: windows.g5.4xlarge.nvidia.gpu
             gpu-arch-type: cuda
-            gpu-arch-version: "11.7"
+            gpu-arch-version: "11.8"
       fail-fast: false
     uses: pytorch/test-infra/.github/workflows/windows_job.yml@main
     with:

gallery/plot_transforms.py

@@ -51,17 +51,23 @@ def plot(imgs, with_orig=True, row_title=None, **imshow_kwargs):
 
 
 ####################################
+# Geometric Transforms
+# --------------------
+# Geometric image transformation refers to the process of altering the geometric properties of an image,
+# such as its shape, size, orientation, or position.
+# It involves applying mathematical operations to the image pixels or coordinates to achieve the desired transformation.
+#
 # Pad
-# ---
+# ~~~
 # The :class:`~torchvision.transforms.Pad` transform
 # (see also :func:`~torchvision.transforms.functional.pad`)
-# fills image borders with some pixel values.
+# pads all image borders with some pixel values.
 padded_imgs = [T.Pad(padding=padding)(orig_img) for padding in (3, 10, 30, 50)]
 plot(padded_imgs)
 
 ####################################
 # Resize
-# ------
+# ~~~~~~
 # The :class:`~torchvision.transforms.Resize` transform
 # (see also :func:`~torchvision.transforms.functional.resize`)
 # resizes an image.
@@ -70,7 +76,7 @@ def plot(imgs, with_orig=True, row_title=None, **imshow_kwargs):
 
 ####################################
 # CenterCrop
-# ----------
+# ~~~~~~~~~~
 # The :class:`~torchvision.transforms.CenterCrop` transform
 # (see also :func:`~torchvision.transforms.functional.center_crop`)
 # crops the given image at the center.
@@ -79,46 +85,13 @@ def plot(imgs, with_orig=True, row_title=None, **imshow_kwargs):
 
 ####################################
 # FiveCrop
-# --------
+# ~~~~~~~~
 # The :class:`~torchvision.transforms.FiveCrop` transform
 # (see also :func:`~torchvision.transforms.functional.five_crop`)
 # crops the given image into four corners and the central crop.
 (top_left, top_right, bottom_left, bottom_right, center) = T.FiveCrop(size=(100, 100))(orig_img)
 plot([top_left, top_right, bottom_left, bottom_right, center])
 
-####################################
-# Grayscale
-# ---------
-# The :class:`~torchvision.transforms.Grayscale` transform
-# (see also :func:`~torchvision.transforms.functional.to_grayscale`)
-# converts an image to grayscale
-gray_img = T.Grayscale()(orig_img)
-plot([gray_img], cmap='gray')
-
-####################################
-# Random transforms
-# -----------------
-# The following transforms are random, which means that the same transfomer
-# instance will produce different result each time it transforms a given image.
-#
-# ColorJitter
-# ~~~~~~~~~~~
-# The :class:`~torchvision.transforms.ColorJitter` transform
-# randomly changes the brightness, saturation, and other properties of an image.
-jitter = T.ColorJitter(brightness=.5, hue=.3)
-jitted_imgs = [jitter(orig_img) for _ in range(4)]
-plot(jitted_imgs)
-
-####################################
-# GaussianBlur
-# ~~~~~~~~~~~~
-# The :class:`~torchvision.transforms.GaussianBlur` transform
-# (see also :func:`~torchvision.transforms.functional.gaussian_blur`)
-# performs gaussian blur transform on an image.
-blurrer = T.GaussianBlur(kernel_size=(5, 9), sigma=(0.1, 5))
-blurred_imgs = [blurrer(orig_img) for _ in range(4)]
-plot(blurred_imgs)
-
 ####################################
 # RandomPerspective
 # ~~~~~~~~~~~~~~~~~
@@ -181,6 +154,45 @@ def plot(imgs, with_orig=True, row_title=None, **imshow_kwargs):
 resized_crops = [resize_cropper(orig_img) for _ in range(4)]
 plot(resized_crops)
 
+####################################
+# Photometric Transforms
+# ----------------------
+# Photometric image transformation refers to the process of modifying the photometric properties of an image,
+# such as its brightness, contrast, color, or tone.
+# These transformations are applied to change the visual appearance of an image
+# while preserving its geometric structure.
+#
+# Except :class:`~torchvision.transforms.Grayscale`, the following transforms are random,
+# which means that the same transform
+# instance will produce different result each time it transforms a given image.
+#
+# Grayscale
+# ~~~~~~~~~
+# The :class:`~torchvision.transforms.Grayscale` transform
+# (see also :func:`~torchvision.transforms.functional.to_grayscale`)
+# converts an image to grayscale
+gray_img = T.Grayscale()(orig_img)
+plot([gray_img], cmap='gray')
+
+####################################
+# ColorJitter
+# ~~~~~~~~~~~
+# The :class:`~torchvision.transforms.ColorJitter` transform
+# randomly changes the brightness, contrast, saturation, hue, and other properties of an image.
+jitter = T.ColorJitter(brightness=.5, hue=.3)
+jitted_imgs = [jitter(orig_img) for _ in range(4)]
+plot(jitted_imgs)
+
+####################################
+# GaussianBlur
+# ~~~~~~~~~~~~
+# The :class:`~torchvision.transforms.GaussianBlur` transform
+# (see also :func:`~torchvision.transforms.functional.gaussian_blur`)
+# performs gaussian blur transform on an image.
+blurrer = T.GaussianBlur(kernel_size=(5, 9), sigma=(0.1, 5))
+blurred_imgs = [blurrer(orig_img) for _ in range(4)]
+plot(blurred_imgs)
+
 ####################################
 # RandomInvert
 # ~~~~~~~~~~~~
@@ -244,6 +256,11 @@ def plot(imgs, with_orig=True, row_title=None, **imshow_kwargs):
 plot(equalized_imgs)
 
 ####################################
+# Augmentation Transforms
+# -----------------------
+# The following transforms are combinations of multiple transforms,
+# either geometric or photometric, or both.
+#
 # AutoAugment
 # ~~~~~~~~~~~
 # The :class:`~torchvision.transforms.AutoAugment` transform
@@ -261,34 +278,36 @@ def plot(imgs, with_orig=True, row_title=None, **imshow_kwargs):
 ####################################
 # RandAugment
 # ~~~~~~~~~~~
-# The :class:`~torchvision.transforms.RandAugment` transform automatically augments the data.
+# The :class:`~torchvision.transforms.RandAugment` is an alternate version of AutoAugment.
 augmenter = T.RandAugment()
 imgs = [augmenter(orig_img) for _ in range(4)]
 plot(imgs)
 
 ####################################
 # TrivialAugmentWide
 # ~~~~~~~~~~~~~~~~~~
-# The :class:`~torchvision.transforms.TrivialAugmentWide` transform automatically augments the data.
+# The :class:`~torchvision.transforms.TrivialAugmentWide` is an alternate implementation of AutoAugment.
+# However, instead of transforming an image multiple times, it transforms an image only once
+# using a random transform from a given list with a random strength number.
 augmenter = T.TrivialAugmentWide()
 imgs = [augmenter(orig_img) for _ in range(4)]
 plot(imgs)
 
 ####################################
 # AugMix
 # ~~~~~~
-# The :class:`~torchvision.transforms.AugMix` transform automatically augments the data.
+# The :class:`~torchvision.transforms.AugMix` transform interpolates between augmented versions of an image.
 augmenter = T.AugMix()
 imgs = [augmenter(orig_img) for _ in range(4)]
 plot(imgs)
 
 ####################################
-# Randomly-applied transforms
+# Randomly-applied Transforms
 # ---------------------------
 #
-# Some transforms are randomly-applied given a probability ``p``.  That is, the
-# transformed image may actually be the same as the original one, even when
-# called with the same transformer instance!
+# The following transforms are randomly-applied given a probability ``p``.  That is, given ``p = 0.5``,
+# there is a 50% chance to return the original image, and a 50% chance to return the transformed image,
+# even when called with the same transform instance!
 #
 # RandomHorizontalFlip
 # ~~~~~~~~~~~~~~~~~~~~

test/test_models.py

@@ -682,10 +682,11 @@ def test_classification_model(model_fn, dev):
     model_name = model_fn.__name__
     if SKIP_BIG_MODEL and is_skippable(model_name, dev):
         pytest.skip("Skipped to reduce memory usage. Set env var SKIP_BIG_MODEL=0 to enable test for this model")
-    if model_name == "vit_h_14" and dev == "cuda":
-        # TODO: investigate why this fail on CI. It doesn't fail on AWS cluster with CUDA 11.6
-        # (can't test with later versions ATM)
-        pytest.xfail("https://github.com/pytorch/vision/issues/7143")
+    if model_name == "resnet101" and dev == "cuda":
+        # TODO: Investigate the Failure with CUDA 11.8: https://github.com/pytorch/vision/issues/7618
+        # TODO: Investigate/followup on previous failure: https://github.com/pytorch/vision/issues/7143
+        # its not happening on CI with CUDA 11.8 anymore. Follow up is needed if its still not resolved.
+        pytest.xfail("https://github.com/pytorch/vision/issues/7618")
     kwargs = {**defaults, **_model_params.get(model_name, {})}
     num_classes = kwargs.get("num_classes")
     input_shape = kwargs.pop("input_shape")