Add missing copyright headers, code linting

Reviewed By: rotabulo

Differential Revision: D32619633

fbshipit-source-id: fd4e66980cb63a46aed071b6ea3849eb5601cc11

inplace_abn/_backend.pyi

@@ -1,32 +1,46 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+
 """Stubs for the native methods"""
 from typing import Tuple, Optional
 
 import torch
 
-
 def statistics(x: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]: ...
-
-
-def reduce_statistics(all_mean: torch.Tensor, all_var: torch.Tensor, all_count: torch.Tensor) \
-        -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]: ...
-
-
-def forward(x: torch.Tensor, mean: torch.Tensor, var: torch.Tensor,
-            weight: Optional[torch.Tensor], bias: Optional[torch.Tensor],
-            eps: float, activation, activation_param: float) -> None: ...
-
-
-def backward_reduce(y_act: torch.Tensor, dy_act: torch.Tensor, weight: Optional[torch.Tensor],
-                    bias: Optional[torch.Tensor], eps: float, activation, activation_param: float) \
-        -> Tuple[torch.Tensor, torch.Tensor]: ...
-
-
-def backward_train(xhat: torch.Tensor, dy: torch.Tensor, var: torch.Tensor, count: torch.Tensor, sum_dy: torch.Tensor,
-                   sum_xhat_dy: torch.Tensor, weight: Optional[torch.Tensor], eps: float) -> torch.Tensor: ...
-
-
-def backward_test(dy: torch.Tensor, var: torch.Tensor, weight: Optional[torch.Tensor], eps: float) -> torch.Tensor: ...
-
+def reduce_statistics(
+    all_mean: torch.Tensor, all_var: torch.Tensor, all_count: torch.Tensor
+) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]: ...
+def forward(
+    x: torch.Tensor,
+    mean: torch.Tensor,
+    var: torch.Tensor,
+    weight: Optional[torch.Tensor],
+    bias: Optional[torch.Tensor],
+    eps: float,
+    activation,
+    activation_param: float,
+) -> None: ...
+def backward_reduce(
+    y_act: torch.Tensor,
+    dy_act: torch.Tensor,
+    weight: Optional[torch.Tensor],
+    bias: Optional[torch.Tensor],
+    eps: float,
+    activation,
+    activation_param: float,
+) -> Tuple[torch.Tensor, torch.Tensor]: ...
+def backward_train(
+    xhat: torch.Tensor,
+    dy: torch.Tensor,
+    var: torch.Tensor,
+    count: torch.Tensor,
+    sum_dy: torch.Tensor,
+    sum_xhat_dy: torch.Tensor,
+    weight: Optional[torch.Tensor],
+    eps: float,
+) -> torch.Tensor: ...
+def backward_test(
+    dy: torch.Tensor, var: torch.Tensor, weight: Optional[torch.Tensor], eps: float
+) -> torch.Tensor: ...
 
 class Activation:
     LeakyReLU = ...

inplace_abn/abn.py

@@ -1,3 +1,5 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+
 from typing import Optional, Any
 
 import torch

inplace_abn/functions.py

@@ -1,3 +1,5 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+
 from typing import Optional, Any
 from warnings import warn
 

inplace_abn/group.py

@@ -1,3 +1,5 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+
 import torch
 import torch.distributed as distributed
 import torch.nn as nn

scripts/dataset/dataset.py

@@ -1,3 +1,5 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+
 import glob
 from itertools import chain
 from os import path
@@ -18,13 +20,15 @@ def __init__(self, in_dir, transform):
 
         # Find all images
         self.images = []
-        for img_path in chain(*(glob.iglob(path.join(self.in_dir, ext)) for ext in SegmentationDataset._EXTENSIONS)):
+        for img_path in chain(
+            *(
+                glob.iglob(path.join(self.in_dir, ext))
+                for ext in SegmentationDataset._EXTENSIONS
+            )
+        ):
             _, name_with_ext = path.split(img_path)
             idx, _ = path.splitext(name_with_ext)
-            self.images.append({
-                "idx": idx,
-                "path": img_path
-            })
+            self.images.append({"idx": idx, "path": img_path})
 
     def __len__(self):
         return len(self.images)

scripts/dataset/sampler.py

@@ -1,3 +1,5 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+
 import torch
 import torch.distributed as dist
 
@@ -29,14 +31,15 @@ def __init__(self, dataset, num_replicas=None, rank=None):
         self.num_replicas = num_replicas
         self.rank = rank
         self.num_samples = (len(self.dataset) // self.num_replicas) + int(
-            (len(self.dataset) % self.num_replicas) < self.rank)
+            (len(self.dataset) % self.num_replicas) < self.rank
+        )
 
     def __iter__(self):
         # deterministically shuffle based on epoch
         indices = torch.arange(0, len(self.dataset))
 
         # subsample
-        indices = indices[self.rank::self.num_replicas]
+        indices = indices[self.rank :: self.num_replicas]
 
         return iter(indices)
 

scripts/dataset/transform.py

@@ -1,3 +1,5 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+
 from PIL import Image
 from torchvision.transforms import functional as tfn
 
@@ -10,8 +12,8 @@ def __init__(self, longest_max_size, rgb_mean, rgb_std):
 
     def __call__(self, img):
         # Scaling
-        scale = self.longest_max_size/float(max(img.size[0],img.size[1]))
-        if scale != 1.:
+        scale = self.longest_max_size / float(max(img.size[0], img.size[1]))
+        if scale != 1.0:
             out_size = tuple(int(dim * scale) for dim in img.size)
             img = img.resize(out_size, resample=Image.BILINEAR)
 

scripts/imagenet/config.py

@@ -1,3 +1,5 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+
 import json
 
 DEFAULTS = {
@@ -10,23 +12,23 @@
         "classes": 1000,
         "dilation": 1,
         "weight_gain_multiplier": 1,  # note: this is ignored if weight_init == kaiming_*
-        "weight_init": "xavier_normal"  # supported: xavier_[normal,uniform], kaiming_[normal,uniform], orthogonal
+        "weight_init": "xavier_normal",  # supported: xavier_[normal,uniform], kaiming_[normal,uniform], orthogonal
     },
     "optimizer": {
         "batch_size": 256,
         "type": "SGD",  # supported: SGD, Adam
         "momentum": 0.9,
         "weight_decay": 1e-4,
-        "clip": 1.,
+        "clip": 1.0,
         "learning_rate": 0.1,
-        "classifier_lr": -1.,  # If -1 use same learning rate as the rest of the network
+        "classifier_lr": -1.0,  # If -1 use same learning rate as the rest of the network
         "nesterov": False,
         "schedule": {
             "type": "constant",  # supported: constant, step, multistep, exponential, linear
             "mode": "epoch",  # supported: epoch, step
             "epochs": 10,
-            "params": {}
-        }
+            "params": {},
+        },
     },
     "input": {
         "scale_train": -1,  # If -1 do not scale
@@ -36,8 +38,8 @@
         "scale_val": 256,  # If -1 do not scale
         "crop_val": 224,
         "mean": [0.485, 0.456, 0.406],
-        "std": [0.229, 0.224, 0.225]
-    }
+        "std": [0.229, 0.224, 0.225],
+    },
 }
 
 

scripts/imagenet/transforms.py

@@ -1,11 +1,17 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+
 from random import sample
 
 import torch
 
 # Default augmentation values compatible with ImageNet data augmentation pipeline
 _DEFAULT_ALPHASTD = 0.1
 _DEFAULT_EIGVAL = [0.2175, 0.0188, 0.0045]
-_DEFAULT_EIGVEC = [[-0.5675, 0.7192, 0.4009], [-0.5808, -0.0045, -0.8140], [-0.5836, -0.6948, 0.4203]]
+_DEFAULT_EIGVEC = [
+    [-0.5675, 0.7192, 0.4009],
+    [-0.5808, -0.0045, -0.8140],
+    [-0.5836, -0.6948, 0.4203],
+]
 _DEFAULT_BCS = [0.4, 0.4, 0.4]
 
 
@@ -19,13 +25,15 @@ def _blend(img1, img2, alpha):
 
 
 class Lighting:
-    def __init__(self, alphastd=_DEFAULT_ALPHASTD, eigval=_DEFAULT_EIGVAL, eigvec=_DEFAULT_EIGVEC):
+    def __init__(
+        self, alphastd=_DEFAULT_ALPHASTD, eigval=_DEFAULT_EIGVAL, eigvec=_DEFAULT_EIGVEC
+    ):
         self._alphastd = alphastd
         self._eigval = eigval
         self._eigvec = eigvec
 
     def __call__(self, img):
-        if self._alphastd == 0.:
+        if self._alphastd == 0.0:
             return img
 
         alpha = torch.normal(img.new_zeros(3), self._alphastd)
@@ -68,7 +76,12 @@ def __call__(self, img):
 
 
 class ColorJitter(object):
-    def __init__(self, saturation=_DEFAULT_BCS[0], brightness=_DEFAULT_BCS[1], contrast=_DEFAULT_BCS[2]):
+    def __init__(
+        self,
+        saturation=_DEFAULT_BCS[0],
+        brightness=_DEFAULT_BCS[1],
+        contrast=_DEFAULT_BCS[2],
+    ):
         self._transforms = []
         if saturation is not None:
             self._transforms.append(Saturation(saturation))