byol_pytorch.py

import copy
import random
from functools import wraps

import torch
from torch import nn
import torch.nn.functional as F

from kornia import augmentation as augs
from kornia import filters, color

# helper functions

def default(val, def_val):
    return def_val if val is None else val

def flatten(t):
    return t.reshape(t.shape[0], -1)

def singleton(cache_key):
    def inner_fn(fn):
        @wraps(fn)
        def wrapper(self, *args, **kwargs):
            instance = getattr(self, cache_key)
            if instance is not None:
                return instance

            instance = fn(self, *args, **kwargs)
            setattr(self, cache_key, instance)
            return instance
        return wrapper
    return inner_fn

# loss fn

def loss_fn(x, y):
    x = F.normalize(x, dim=-1, p=2)
    y = F.normalize(y, dim=-1, p=2)
    return 2 - 2 * (x * y).sum(dim=-1)

# augmentation utils

class RandomApply(nn.Module):
    def __init__(self, fn, p):
        super().__init__()
        self.fn = fn
        self.p = p
    def forward(self, x):
        if random.random() > self.p:
            return x
        return self.fn(x)

# exponential moving average

class EMA():
    def __init__(self, beta):
        super().__init__()
        self.beta = beta

    def update_average(self, old, new):
        if old is None:
            return new
        return old * self.beta + (1 - self.beta) * new

def update_moving_average(ema_updater, ma_model, current_model):
    for current_params, ma_params in zip(current_model.parameters(), ma_model.parameters()):
        old_weight, up_weight = ma_params.data, current_params.data
        ma_params.data = ema_updater.update_average(old_weight, up_weight)

# MLP class for projector and predictor

class MLP(nn.Module):
    def __init__(self, dim, projection_size, hidden_size = 4096):
        super().__init__()
        self.net = nn.Sequential(
            nn.Linear(dim, hidden_size),
            nn.BatchNorm1d(hidden_size),
            nn.ReLU(inplace=True),
            nn.Linear(hidden_size, projection_size)
        )

    def forward(self, x):
        return self.net(x)

# a wrapper class for the base neural network
# will manage the interception of the hidden layer output
# and pipe it into the projecter and predictor nets

class NetWrapper(nn.Module):
    def __init__(self, net, projection_size, projection_hidden_size, layer = -2):
        super().__init__()
        self.net = net
        self.layer = layer

        self.projector = None
        self.projection_size = projection_size
        self.projection_hidden_size = projection_hidden_size

        self.hidden = None
        self.hook_registered = False

    def _find_layer(self):
        if type(self.layer) == str:
            modules = dict([*self.net.named_modules()])
            return modules.get(self.layer, None)
        elif type(self.layer) == int:
            children = [*self.net.children()]
            return children[self.layer]
        return None

    def _hook(self, _, __, output):
        self.hidden = flatten(output)

    def _register_hook(self):
        layer = self._find_layer()
        assert layer is not None, f'hidden layer ({self.layer}) not found'
        handle = layer.register_forward_hook(self._hook)
        self.hook_registered = True

    @singleton('projector')
    def _get_projector(self, hidden):
        _, dim = hidden.shape
        projector = MLP(dim, self.projection_size, self.projection_hidden_size)
        return projector.to(hidden)

    def get_representation(self, x):
        if self.layer == -1:
            return self.net(x)

        if not self.hook_registered:
            self._register_hook()

        _ = self.net(x)
        hidden = self.hidden
        self.hidden = None
        assert hidden is not None, f'hidden layer {self.layer} never emitted an output'
        return hidden

    def forward(self, x):
        representation = self.get_representation(x)
        projector = self._get_projector(representation)
        projection = projector(representation)
        return projection

# main class

class BYOL(nn.Module):
    def __init__(self, net, image_size, hidden_layer = -2, projection_size = 256, projection_hidden_size = 4096, augment_fn = None, moving_average_decay = 0.99):
        super().__init__()

        # default SimCLR augmentation

        DEFAULT_AUG = nn.Sequential(
            RandomApply(augs.ColorJitter(0.8, 0.8, 0.8, 0.2), p=0.8),
            augs.RandomGrayscale(p=0.2),
            augs.RandomHorizontalFlip(),
            RandomApply(filters.GaussianBlur2d((3, 3), (1.5, 1.5)), p=0.1),
            augs.RandomResizedCrop((image_size, image_size)),
            color.Normalize(mean=torch.tensor([0.485, 0.456, 0.406]), std=torch.tensor([0.229, 0.224, 0.225]))
        )

        self.augment = default(augment_fn, DEFAULT_AUG)

        self.online_encoder = NetWrapper(net, projection_size, projection_hidden_size, layer=hidden_layer)
        self.target_encoder = None
        self.target_ema_updater = EMA(moving_average_decay)

        self.online_predictor = MLP(projection_size, projection_size, projection_hidden_size)

        # send a mock image tensor to instantiate singleton parameters
        self.forward(torch.randn(2, 3, image_size, image_size))

    @singleton('target_encoder')
    def _get_target_encoder(self):
        target_encoder = copy.deepcopy(self.online_encoder)
        return target_encoder

    def reset_moving_average(self):
        del self.target_encoder
        self.target_encoder = None

    def update_moving_average(self):
        assert self.target_encoder is not None, 'target encoder has not been created yet'
        update_moving_average(self.target_ema_updater, self.target_encoder, self.online_encoder)

    def forward(self, x):
        image_one, image_two = self.augment(x), self.augment(x)

        online_proj_one = self.online_encoder(image_one)
        online_proj_two = self.online_encoder(image_two)

        online_pred_one = self.online_predictor(online_proj_one)
        online_pred_two = self.online_predictor(online_proj_two)

        with torch.no_grad():
            target_encoder = self._get_target_encoder()
            target_proj_one = target_encoder(image_one)
            target_proj_two = target_encoder(image_two)

        loss_one = loss_fn(online_pred_one, target_proj_two.detach())
        loss_two = loss_fn(online_pred_two, target_proj_one.detach())

        loss = loss_one + loss_two
        return loss.mean()