Fix some attributions, add copyrights to some file docstrings

avg_checkpoints.py

@@ -9,7 +9,7 @@ For any hope of decent results, the checkpoints should be from the same or child
 EMA (exponential moving average) of the model weights or performing SWA (stochastic
 weight averaging), but post-training.
 
-Hacked together by Ross Wightman (https://github.com/rwightman)
+Hacked together by / Copyright 2020 Ross Wightman (https://github.com/rwightman)
 """
 import torch
 import argparse

clean_checkpoint.py

@@ -5,7 +5,7 @@ Takes training checkpoints with GPU tensors, optimizer state, extra dict keys, e
 and outputs a CPU  tensor checkpoint with only the `state_dict` along with SHA256
 calculation for model zoo compatibility.
 
-Hacked together by Ross Wightman (https://github.com/rwightman)
+Hacked together by / Copyright 2020 Ross Wightman (https://github.com/rwightman)
 """
 import torch
 import argparse

inference.py

@@ -3,7 +3,7 @@
 
 An example inference script that outputs top-k class ids for images in a folder into a csv.
 
-Hacked together by Ross Wightman (https://github.com/rwightman)
+Hacked together by / Copyright 2020 Ross Wightman (https://github.com/rwightman)
 """
 import os
 import time

timm/data/auto_augment.py

@@ -15,7 +15,7 @@ Papers:
     RandAugment: Practical automated data augmentation... - https://arxiv.org/abs/1909.13719
     AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty - https://arxiv.org/abs/1912.02781
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 import random
 import math

timm/data/dataset.py

@@ -1,3 +1,7 @@
+""" Quick n Simple Image Folder, Tarfile based DataSet
+
+Hacked together by / Copyright 2020 Ross Wightman
+"""
 from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function

timm/data/loader.py

@@ -1,3 +1,11 @@
+""" Loader Factory, Fast Collate, CUDA Prefetcher
+
+Prefetcher and Fast Collate inspired by NVIDIA APEX example at
+https://github.com/NVIDIA/apex/commit/d5e2bb4bdeedd27b1dfaf5bb2b24d6c000dee9be#diff-cf86c282ff7fba81fad27a559379d5bf
+
+Hacked together by / Copyright 2020 Ross Wightman
+"""
+
 import torch.utils.data
 import numpy as np
 

timm/data/mixup.py

@@ -1,3 +1,8 @@
+""" Mixup
+Paper: `mixup: Beyond Empirical Risk Minimization` - https://arxiv.org/abs/1710.09412
+
+Hacked together by / Copyright 2020 Ross Wightman
+"""
 import numpy as np
 import torch
 

timm/data/random_erasing.py

@@ -1,3 +1,10 @@
+""" Random Erasing (Cutout)
+
+Originally inspired by impl at https://github.com/zhunzhong07/Random-Erasing, Apache 2.0
+Copyright Zhun Zhong & Liang Zheng
+
+Hacked together by / Copyright 2020 Ross Wightman
+"""
 import random
 import math
 import torch

timm/data/real_labels.py

@@ -1,3 +1,9 @@
+""" Real labels evaluator for ImageNet
+Paper: `Are we done with ImageNet?` - https://arxiv.org/abs/2006.07159
+Based on Numpy example at https://github.com/google-research/reassessed-imagenet
+
+Hacked together by / Copyright 2020 Ross Wightman
+"""
 import os
 import json
 import numpy as np

timm/data/tf_preprocessing.py

@@ -1,3 +1,12 @@
+""" Tensorflow Preprocessing Adapter
+
+Allows use of Tensorflow preprocessing pipeline in PyTorch Transform
+
+Copyright of original Tensorflow code below.
+
+Hacked together by / Copyright 2020 Ross Wightman
+"""
+
 # Copyright 2018 The TensorFlow Authors. All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

timm/data/transforms_factory.py

@@ -1,5 +1,7 @@
 """ Transforms Factory
 Factory methods for building image transforms for use with TIMM (PyTorch Image Models)
+
+Hacked together by / Copyright 2020 Ross Wightman
 """
 import math
 

timm/loss/jsd.py

@@ -12,7 +12,7 @@ class JsdCrossEntropy(nn.Module):
     From paper: 'AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty -
     https://arxiv.org/abs/1912.02781
 
-    Hacked together by Ross Wightman
+    Hacked together by / Copyright 2020 Ross Wightman
     """
     def __init__(self, num_splits=3, alpha=12, smoothing=0.1):
         super().__init__()

timm/models/dpn.py

@@ -2,8 +2,9 @@
 Based on original MXNet implementation https://github.com/cypw/DPNs with
 many ideas from another PyTorch implementation https://github.com/oyam/pytorch-DPNs.
 
-This implementation is compatible with the pretrained weights
+This implementation is compatible with the pretrained weights from cypw's MXNet implementation.
-from cypw's MXNet implementation.
+
+Hacked together by / Copyright 2020 Ross Wightman
 """
 from __future__ import absolute_import
 from __future__ import division

timm/models/efficientnet.py

@@ -22,7 +22,7 @@ An implementation of EfficienNet that covers variety of related models with effi
 
 * And likely more...
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 import torch
 import torch.nn as nn

timm/models/efficientnet_blocks.py

@@ -1,3 +1,8 @@
+""" EfficientNet, MobileNetV3, etc Blocks
+
+Hacked together by / Copyright 2020 Ross Wightman
+"""
+
 import torch
 import torch.nn as nn
 from torch.nn import functional as F

timm/models/efficientnet_builder.py

@@ -1,3 +1,11 @@
+""" EfficientNet, MobileNetV3, etc Builder
+
+Assembles EfficieNet and related network feature blocks from string definitions.
+Handles stride, dilation calculations, and selects feature extraction points.
+
+Hacked together by / Copyright 2020 Ross Wightman
+"""
+
 import logging
 import math
 import re

timm/models/features.py

@@ -3,7 +3,10 @@
 A collection of classes, functions, modules to help extract features from models
 and provide a common interface for describing them.
 
-Hacked together by Ross Wightman
+The return_layers, module re-writing idea inspired by torchvision IntermediateLayerGetter
+https://github.com/pytorch/vision/blob/d88d8961ae51507d0cb680329d985b1488b1b76b/torchvision/models/_utils.py
+
+Hacked together by / Copyright 2020 Ross Wightman
 """
 from collections import OrderedDict, defaultdict
 from copy import deepcopy

timm/models/gluon_xception.py

@@ -4,7 +4,7 @@ This is a port of the Gluon Xception code and weights, itself ported from a PyTo
 Gluon model: (https://gluon-cv.mxnet.io/_modules/gluoncv/model_zoo/xception.html)
 Original PyTorch DeepLab impl: https://github.com/jfzhang95/pytorch-deeplab-xception
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 from collections import OrderedDict
 

timm/models/inception_v3.py

@@ -1,3 +1,8 @@
+""" Inception-V3
+
+Originally from torchvision Inception3 model
+Licensed BSD-Clause 3 https://github.com/pytorch/vision/blob/master/LICENSE
+"""
 import torch
 import torch.nn as nn
 import torch.nn.functional as F

timm/models/layers/activations.py

@@ -3,7 +3,7 @@
 A collection of activations fn and modules with a common interface so that they can
 easily be swapped. All have an `inplace` arg even if not used.
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 
 import torch

timm/models/layers/activations_jit.py

@@ -7,7 +7,7 @@ All jit scripted activations are lacking in-place variations on purpose, scripte
 currently work across in-place op boundaries, thus performance is equal to or less than the non-scripted
 versions if they contain in-place ops.
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 
 import torch

timm/models/layers/activations_me.py

@@ -6,7 +6,7 @@ easily be swapped. All have an `inplace` arg even if not used.
 These activations are not compatible with jit scripting or ONNX export of the model, please use either
 the JIT or basic versions of the activations.
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 
 import torch

timm/models/layers/adaptive_avgmax_pool.py

@@ -7,7 +7,7 @@ Adaptive pooling with the ability to select the type of pooling from:
 
 Both a functional and a nn.Module version of the pooling is provided.
 
-Author: Ross Wightman (rwightman)
+Hacked together by / Copyright 2020 Ross Wightman
 """
 import torch
 import torch.nn as nn

timm/models/layers/cbam.py

@@ -5,7 +5,7 @@ Experimental impl of CBAM: Convolutional Block Attention Module: https://arxiv.o
 WARNING: Results with these attention layers have been mixed. They can significantly reduce performance on
 some tasks, especially fine-grained it seems. I may end up removing this impl.
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 
 import torch

timm/models/layers/cond_conv2d.py

@@ -3,7 +3,7 @@
 Paper: CondConv: Conditionally Parameterized Convolutions for Efficient Inference
 (https://arxiv.org/abs/1904.04971)
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 
 import math

timm/models/layers/conv2d_same.py

@@ -1,6 +1,6 @@
 """ Conv2d w/ Same Padding
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 import torch
 import torch.nn as nn

timm/models/layers/conv_bn_act.py

@@ -1,6 +1,6 @@
 """ Conv2d + BN + Act
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 from torch import nn as nn
 

timm/models/layers/create_act.py

@@ -1,3 +1,6 @@
+""" Activation Factory
+Hacked together by / Copyright 2020 Ross Wightman
+"""
 from .activations import *
 from .activations_jit import *
 from .activations_me import *

timm/models/layers/create_attn.py

@@ -1,6 +1,6 @@
 """ Select AttentionFactory Method
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 import torch
 from .se import SEModule, EffectiveSEModule

timm/models/layers/create_conv2d.py

@@ -1,6 +1,6 @@
 """ Create Conv2d Factory Method
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 
 from .mixed_conv2d import MixedConv2d

timm/models/layers/create_norm_act.py

@@ -1,3 +1,11 @@
+""" NormAct (Normalizaiton + Activation Layer) Factory
+
+Create norm + act combo modules that attempt to be backwards compatible with separate norm + act
+isntances in models. Where these are used it will be possible to swap separate BN + act layers with
+combined modules like IABN or EvoNorms.
+
+Hacked together by / Copyright 2020 Ross Wightman
+"""
 import types
 import functools
 

timm/models/layers/drop.py

@@ -12,7 +12,7 @@ DropBlock impl inspired by two Tensorflow impl that I liked:
  - https://github.com/tensorflow/tpu/blob/master/models/official/resnet/resnet_model.py#L74
  - https://github.com/clovaai/assembled-cnn/blob/master/nets/blocks.py
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 import torch
 import torch.nn as nn

timm/models/layers/evo_norm.py

@@ -6,7 +6,7 @@ in terms of memory usage and throughput (roughly 5x mem, 1/2 - 1/3x speed).
 
 Still very much a WIP, fiddling with buffer usage, in-place/jit optimizations, and layouts.
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 
 import torch

timm/models/layers/helpers.py

@@ -1,6 +1,6 @@
 """ Layer/Module Helpers
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 from itertools import repeat
 from torch._six import container_abcs

timm/models/layers/median_pool.py

@@ -1,8 +1,9 @@
-import math
+""" Median Pool
-import torch
+Hacked together by / Copyright 2020 Ross Wightman
+"""
 import torch.nn as nn
 import torch.nn.functional as F
-from torch.nn.modules.utils import _pair, _quadruple
+from .helpers import tup_pair, tup_quadruple
 
 
 class MedianPool2d(nn.Module):
@@ -16,9 +17,9 @@ class MedianPool2d(nn.Module):
     """
     def __init__(self, kernel_size=3, stride=1, padding=0, same=False):
         super(MedianPool2d, self).__init__()
-        self.k = _pair(kernel_size)
+        self.k = tup_pair(kernel_size)
-        self.stride = _pair(stride)
+        self.stride = tup_pair(stride)
-        self.padding = _quadruple(padding)  # convert to l, r, t, b
+        self.padding = tup_quadruple(padding)  # convert to l, r, t, b
         self.same = same
 
     def _padding(self, x):

timm/models/layers/mixed_conv2d.py

@@ -2,7 +2,7 @@
 
 Paper: MixConv: Mixed Depthwise Convolutional Kernels (https://arxiv.org/abs/1907.09595)
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 
 import torch

timm/models/layers/padding.py

@@ -1,6 +1,6 @@
 """ Padding Helpers
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 import math
 from typing import List, Tuple

timm/models/layers/pool2d_same.py

@@ -1,6 +1,6 @@
 """ AvgPool2d w/ Same Padding
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 import torch
 import torch.nn as nn

timm/models/layers/selective_kernel.py

@@ -2,7 +2,7 @@
 
 Paper: Selective Kernel Networks (https://arxiv.org/abs/1903.06586)
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 import torch
 from torch import nn as nn

timm/models/layers/separable_conv.py

@@ -1,3 +1,10 @@
+""" Depthwise Separable Conv Modules
+
+Basic DWS convs. Other variations of DWS exist with batch norm or activations between the
+DW and PW convs such as the Depthwise modules in MobileNetV2 / EfficientNet and Xception.
+
+Hacked together by / Copyright 2020 Ross Wightman
+"""
 from torch import nn as nn
 
 from .create_conv2d import create_conv2d

timm/models/layers/split_batchnorm.py

@@ -9,7 +9,7 @@ This allows easily removing the auxiliary BN layers after training to efficientl
 achieve the 'Auxiliary BatchNorm' as described in the AdvProp Paper, section 4.2,
 'Disentangled Learning via An Auxiliary BN'
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 import torch
 import torch.nn as nn

timm/models/layers/test_time_pool.py

@@ -1,6 +1,6 @@
 """ Test Time Pooling (Average-Max Pool)
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 
 import logging

timm/models/mobilenetv3.py

@@ -5,7 +5,7 @@ A PyTorch impl of MobileNet-V3, compatible with TF weights from official impl.
 
 Paper: Searching for MobileNetV3 - https://arxiv.org/abs/1905.02244
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 import torch
 import torch.nn as nn

timm/models/registry.py

@@ -1,3 +1,7 @@
+""" Model Registry
+Hacked together by / Copyright 2020 Ross Wightman
+"""
+
 import sys
 import re
 import fnmatch

timm/models/regnet.py

@@ -11,6 +11,7 @@ Weights from original impl have been modified
 * removed training specific dict entries from checkpoints and keep model state_dict only
 * remap names to match the ones here
 
+Hacked together by / Copyright 2020 Ross Wightman
 """
 import numpy as np
 import torch.nn as nn

timm/models/resnet.py

@@ -4,6 +4,7 @@ This started as a copy of https://github.com/pytorch/vision 'resnet.py' (BSD-3-C
 additional dropout and dynamic global avg/max pool.
 
 ResNeXt, SE-ResNeXt, SENet, and MXNet Gluon stem/downsample variants, tiered stems added by Ross Wightman
+Copyright 2020 Ross Wightman
 """
 import math
 import copy

timm/models/rexnet.py

@@ -7,6 +7,7 @@ Adapted from original impl at https://github.com/clovaai/rexnet
 Copyright (c) 2020-present NAVER Corp. MIT license
 
 Changes for timm, feature extraction, and rounded channel variant hacked together by Ross Wightman
+Copyright 2020 Ross Wightman
 """
 
 import torch.nn as nn

timm/models/sknet.py

@@ -6,7 +6,7 @@ This was inspired by reading 'Compounding the Performance Improvements...' (http
 and a streamlined impl at https://github.com/clovaai/assembled-cnn but I ended up building something closer
 to the original paper with some modifications of my own to better balance param count vs accuracy.
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 import math
 

timm/models/vovnet.py

@@ -8,7 +8,7 @@ Looked at  https://github.com/youngwanLEE/vovnet-detectron2 &
 https://github.com/stigma0617/VoVNet.pytorch/blob/master/models_vovnet/vovnet.py
 for some reference, rewrote most of the code.
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 
 from typing import List

timm/models/xception_aligned.py

@@ -3,7 +3,7 @@
 This is a correct, from scratch impl of Aligned Xception (Deeplab) models compatible with TF weights at
 https://github.com/tensorflow/models/blob/master/research/deeplab/g3doc/model_zoo.md
 
-Hacked together by Ross Wightman
+Hacked together by / Copyright 2020 Ross Wightman
 """
 from collections import OrderedDict
 

timm/optim/lookahead.py

@@ -1,6 +1,8 @@
 """ Lookahead Optimizer Wrapper.
 Implementation modified from: https://github.com/alphadl/lookahead.pytorch
 Paper: `Lookahead Optimizer: k steps forward, 1 step back` - https://arxiv.org/abs/1907.08610
+
+Hacked together by / Copyright 2020 Ross Wightman
 """
 import torch
 from torch.optim.optimizer import Optimizer

timm/optim/optim_factory.py

@@ -1,3 +1,6 @@
+""" Optimizer Factory w/ Custom Weight Decay
+Hacked together by / Copyright 2020 Ross Wightman
+"""
 import torch
 from torch import optim as optim
 from timm.optim import Nadam, RMSpropTF, AdamW, RAdam, NovoGrad, NvNovoGrad, Lookahead

timm/optim/rmsprop_tf.py

@@ -1,3 +1,12 @@
+""" RMSProp modified to behave like Tensorflow impl
+
+Originally cut & paste from PyTorch RMSProp
+https://github.com/pytorch/pytorch/blob/063946d2b3f3f1e953a2a3b54e0b34f1393de295/torch/optim/rmsprop.py
+Licensed under BSD-Clause 3 (ish), https://github.com/pytorch/pytorch/blob/master/LICENSE
+
+Modifications Copyright 2020 Ross Wightman
+"""
+
 import torch
 from torch.optim import Optimizer
 
@@ -6,7 +15,12 @@ class RMSpropTF(Optimizer):
     """Implements RMSprop algorithm (TensorFlow style epsilon)
 
     NOTE: This is a direct cut-and-paste of PyTorch RMSprop with eps applied before sqrt
-    to closer match Tensorflow for matching hyper-params.
+    and a few other modifications to closer match Tensorflow for matching hyper-params.
+
+    Noteworthy changes include:
+    1. Epsilon applied inside square-root
+    2. square_avg initialized to ones
+    3. LR scaling of update accumulated in momentum buffer
 
     Proposed by G. Hinton in his
     `course <http://www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf>`_.

timm/scheduler/cosine_lr.py

@@ -1,3 +1,9 @@
+""" Cosine Scheduler
+
+Cosine LR schedule with warmup, cycle/restarts, noise.
+
+Hacked together by / Copyright 2020 Ross Wightman
+"""
 import logging
 import math
 import numpy as np

timm/scheduler/plateau_lr.py

@@ -1,3 +1,9 @@
+""" Plateau Scheduler
+
+Adapts PyTorch plateau scheduler and allows application of noise, warmup.
+
+Hacked together by / Copyright 2020 Ross Wightman
+"""
 import torch
 
 from .scheduler import Scheduler

timm/scheduler/scheduler_factory.py

@@ -1,3 +1,6 @@
+""" Scheduler Factory
+Hacked together by / Copyright 2020 Ross Wightman
+"""
 from .cosine_lr import CosineLRScheduler
 from .tanh_lr import TanhLRScheduler
 from .step_lr import StepLRScheduler

timm/scheduler/step_lr.py

@@ -1,3 +1,9 @@
+""" Step Scheduler
+
+Basic step LR schedule with warmup, noise.
+
+Hacked together by / Copyright 2020 Ross Wightman
+"""
 import math
 import torch
 

timm/scheduler/tanh_lr.py

@@ -1,3 +1,9 @@
+""" TanH Scheduler
+
+TanH schedule with warmup, cycle/restarts, noise.
+
+Hacked together by / Copyright 2020 Ross Wightman
+"""
 import logging
 import math
 import numpy as np

timm/utils.py

@@ -1,3 +1,8 @@
+""" Common training and validation utilities
+
+Hacked together by / Copyright 2020 Ross Wightman
+"""
+
 from copy import deepcopy
 
 import torch

train.py

@@ -12,7 +12,7 @@ This script was started from an early version of the PyTorch ImageNet example
 NVIDIA CUDA specific speedups adopted from NVIDIA Apex examples
 (https://github.com/NVIDIA/apex/tree/master/examples/imagenet)
 
-Hacked together by Ross Wightman (https://github.com/rwightman)
+Hacked together by / Copyright 2020 Ross Wightman (https://github.com/rwightman)
 """
 import argparse
 import time

validate.py

@@ -5,7 +5,7 @@ This is intended to be a lean and easily modifiable ImageNet validation script f
 models or training checkpoints against ImageNet or similarly organized image datasets. It prioritizes
 canonical PyTorch, standard Python style, and good performance. Repurpose as you see fit.
 
-Hacked together by Ross Wightman (https://github.com/rwightman)
+Hacked together by / Copyright 2020 Ross Wightman (https://github.com/rwightman)
 """
 import argparse
 import os