kivantium活動日記

プログラムを使っていろいろやります

PyTorchでファインチューニングしたモデルをONNXで利用する

昨日の作業の結果、Illustration2Vecのモデルが大きすぎて貧弱なサーバーでは使えないことが分かりました。今のところ二次元画像判別器の特徴量抽出にしか使っていないので、もっと軽いモデルでも代用できるはずです。軽いモデルとして有名なSqueezenetをこれまで集めたデータでファインチューニングして様子を見てみることにします。

ファインチューニングとONNXへのエキスポート

PyTorchのチュートリアルが丁寧に説明してくれているので、これをコピペして継ぎ接ぎするだけです。

継ぎ接ぎしたものがこちらになります。これを実行するとmodel.onnxというファイルが作成されます。 ONNX版Illustration2Vecのモデルサイズが910Mに対して、このモデルは2.8MBなのでだいぶ小さくなりました。精度もだいたい同じくらいだと思います。

from __future__ import print_function
from __future__ import division
import torch
import torch.nn as nn
import torch.optim as optim
import numpy as np
import torchvision
from torchvision import datasets, models, transforms
import matplotlib.pyplot as plt
import time
import os
import copy

print("PyTorch Version: ",torch.__version__)
print("Torchvision Version: ",torchvision.__version__)

# Top level data directory. Here we assume the format of the directory conforms
# to the ImageFolder structure
data_dir = "./data/images"

# Models to choose from [resnet, alexnet, vgg, squeezenet, densenet, inception]
model_name = "squeezenet"

# Number of classes in the dataset
num_classes = 2

# Batch size for training (change depending on how much memory you have)
batch_size = 8

# Number of epochs to train for
num_epochs = 1

# Flag for feature extracting. When False, we finetune the whole model,
#   when True we only update the reshaped layer params
feature_extract = True

def train_model(model, dataloaders, criterion, optimizer, num_epochs=25, is_inception=False):
    since = time.time()

    val_acc_history = []

    best_model_wts = copy.deepcopy(model.state_dict())
    best_acc = 0.0

    for epoch in range(num_epochs):
        print('Epoch {}/{}'.format(epoch, num_epochs - 1))
        print('-' * 10)

        # Each epoch has a training and validation phase
        for phase in ['train', 'val']:
            if phase == 'train':
                model.train()  # Set model to training mode
            else:
                model.eval()   # Set model to evaluate mode

            running_loss = 0.0
            running_corrects = 0

            # Iterate over data.
            for inputs, labels in dataloaders[phase]:
                inputs = inputs.to(device)
                labels = labels.to(device)

                # zero the parameter gradients
                optimizer.zero_grad()

                # forward
                # track history if only in train
                with torch.set_grad_enabled(phase == 'train'):
                    # Get model outputs and calculate loss
                    # Special case for inception because in training it has an auxiliary output. In train
                    #   mode we calculate the loss by summing the final output and the auxiliary output
                    #   but in testing we only consider the final output.
                    if is_inception and phase == 'train':
                        # From https://discuss.pytorch.org/t/how-to-optimize-inception-model-with-auxiliary-classifiers/7958
                        outputs, aux_outputs = model(inputs)
                        loss1 = criterion(outputs, labels)
                        loss2 = criterion(aux_outputs, labels)
                        loss = loss1 + 0.4*loss2
                    else:
                        outputs = model(inputs)
                        loss = criterion(outputs, labels)

                    _, preds = torch.max(outputs, 1)

                    # backward + optimize only if in training phase
                    if phase == 'train':
                        loss.backward()
                        optimizer.step()

                # statistics
                running_loss += loss.item() * inputs.size(0)
                running_corrects += torch.sum(preds == labels.data)

            epoch_loss = running_loss / len(dataloaders[phase].dataset)
            epoch_acc = running_corrects.double() / len(dataloaders[phase].dataset)

            print('{} Loss: {:.4f} Acc: {:.4f}'.format(phase, epoch_loss, epoch_acc))

            # deep copy the model
            if phase == 'val' and epoch_acc > best_acc:
                best_acc = epoch_acc
                best_model_wts = copy.deepcopy(model.state_dict())
            if phase == 'val':
                val_acc_history.append(epoch_acc)

        print()

    time_elapsed = time.time() - since
    print('Training complete in {:.0f}m {:.0f}s'.format(time_elapsed // 60, time_elapsed % 60))
    print('Best val Acc: {:4f}'.format(best_acc))

    # load best model weights
    model.load_state_dict(best_model_wts)
    return model, val_acc_history

def set_parameter_requires_grad(model, feature_extracting):
    if feature_extracting:
        for param in model.parameters():
            param.requires_grad = False

def initialize_model(model_name, num_classes, feature_extract, use_pretrained=True):
    # Initialize these variables which will be set in this if statement. Each of these
    #   variables is model specific.
    model_ft = None
    input_size = 0

    if model_name == "resnet":
        """ Resnet18
        """
        model_ft = models.resnet18(pretrained=use_pretrained)
        set_parameter_requires_grad(model_ft, feature_extract)
        num_ftrs = model_ft.fc.in_features
        model_ft.fc = nn.Linear(num_ftrs, num_classes)
        input_size = 224

    elif model_name == "alexnet":
        """ Alexnet
        """
        model_ft = models.alexnet(pretrained=use_pretrained)
        set_parameter_requires_grad(model_ft, feature_extract)
        num_ftrs = model_ft.classifier[6].in_features
        model_ft.classifier[6] = nn.Linear(num_ftrs,num_classes)
        input_size = 224

    elif model_name == "vgg":
        """ VGG11_bn
        """
        model_ft = models.vgg11_bn(pretrained=use_pretrained)
        set_parameter_requires_grad(model_ft, feature_extract)
        num_ftrs = model_ft.classifier[6].in_features
        model_ft.classifier[6] = nn.Linear(num_ftrs,num_classes)
        input_size = 224

    elif model_name == "squeezenet":
        """ Squeezenet
        """
        model_ft = models.squeezenet1_0(pretrained=use_pretrained)
        set_parameter_requires_grad(model_ft, feature_extract)
        model_ft.classifier[1] = nn.Conv2d(512, num_classes, kernel_size=(1,1), stride=(1,1))
        model_ft.num_classes = num_classes
        input_size = 224

    elif model_name == "densenet":
        """ Densenet
        """
        model_ft = models.densenet121(pretrained=use_pretrained)
        set_parameter_requires_grad(model_ft, feature_extract)
        num_ftrs = model_ft.classifier.in_features
        model_ft.classifier = nn.Linear(num_ftrs, num_classes)
        input_size = 224

    elif model_name == "inception":
        """ Inception v3
        Be careful, expects (299,299) sized images and has auxiliary output
        """
        model_ft = models.inception_v3(pretrained=use_pretrained)
        set_parameter_requires_grad(model_ft, feature_extract)
        # Handle the auxilary net
        num_ftrs = model_ft.AuxLogits.fc.in_features
        model_ft.AuxLogits.fc = nn.Linear(num_ftrs, num_classes)
        # Handle the primary net
        num_ftrs = model_ft.fc.in_features
        model_ft.fc = nn.Linear(num_ftrs,num_classes)
        input_size = 299

    else:
        print("Invalid model name, exiting...")
        exit()

    return model_ft, input_size

# Initialize the model for this run
model_ft, input_size = initialize_model(model_name, num_classes, feature_extract, use_pretrained=True)

# Print the model we just instantiated
print(model_ft)

# Data augmentation and normalization for training
# Just normalization for validation
data_transforms = {
    'train': transforms.Compose([
        transforms.RandomResizedCrop(input_size),
        transforms.RandomHorizontalFlip(),
        transforms.ToTensor(),
        transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
    ]),
    'val': transforms.Compose([
        transforms.Resize(input_size),
        transforms.CenterCrop(input_size),
        transforms.ToTensor(),
        transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
    ]),
}

print("Initializing Datasets and Dataloaders...")

# Create training and validation datasets
image_datasets = {x: datasets.ImageFolder(os.path.join(data_dir, x), data_transforms[x]) for x in ['train', 'val']}
# Create training and validation dataloaders
dataloaders_dict = {x: torch.utils.data.DataLoader(image_datasets[x], batch_size=batch_size, shuffle=True, num_workers=4) for x in ['train', 'val']}

# Detect if we have a GPU available
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

model_ft = model_ft.to(device)

# Gather the parameters to be optimized/updated in this run. If we are
#  finetuning we will be updating all parameters. However, if we are
#  doing feature extract method, we will only update the parameters
#  that we have just initialized, i.e. the parameters with requires_grad
#  is True.
params_to_update = model_ft.parameters()
print("Params to learn:")
if feature_extract:
    params_to_update = []
    for name,param in model_ft.named_parameters():
        if param.requires_grad == True:
            params_to_update.append(param)
            print("\t",name)
else:
    for name,param in model_ft.named_parameters():
        if param.requires_grad == True:
            print("\t",name)

# Observe that all parameters are being optimized
optimizer_ft = optim.SGD(params_to_update, lr=0.001, momentum=0.9)

# Setup the loss fxn
criterion = nn.CrossEntropyLoss()

# Train and evaluate
model_ft, hist = train_model(model_ft, dataloaders_dict, criterion, optimizer_ft, num_epochs=num_epochs, is_inception=(model_name=="inception"))

# Save PyTorch model to file
torch.save(model_ft.to('cpu').state_dict(), 'model.pth')

# Input to the model
x = torch.randn(1, 3, 224, 224, requires_grad=True)
torch_out = model_ft(x)

# Export the model
torch.onnx.export(model_ft,               # model being run
                  x,                         # model input (or a tuple for multiple inputs)
                  "model.onnx",                # where to save the model (can be a file or file-like object)
                  export_params=True,        # store the trained parameter weights inside the model file
                  opset_version=10,          # the ONNX version to export the model to
                  do_constant_folding=True,  # whether to execute constant folding for optimization
                  input_names = ['input'],   # the model's input names
                  output_names = ['output'], # the model's output names
                  dynamic_axes={'input' : {0 : 'batch_size'},    # variable lenght axes
                                'output' : {0 : 'batch_size'}})

ONNXモデルの利用

こうして作成したONNXモデルをPyTorchを使わずに利用するコードはこんな感じです。

import os
from PIL import Image

import numpy as np
import onnxruntime

# 中心を正方形に切り抜いてリサイズ
def crop_and_resize(img, size):
    width, height = img.size
    crop_size = min(width, height)
    img_crop = img.crop(((width - crop_size) // 2, (height - crop_size) // 2,
                         (width + crop_size) // 2, (height + crop_size) // 2))
    return img_crop.resize((size, size))

img_mean = np.asarray([0.485, 0.456, 0.406])
img_std = np.asarray([0.229, 0.224, 0.225])

ort_session = onnxruntime.InferenceSession(
    os.path.join(os.path.dirname(__file__), "model.onnx"))

img = Image.open('image.jpg').convert('RGB')
img = crop_and_resize(img, 224)

# 画像の正規化
img_np = np.asarray(img).astype(np.float32)/255.0
img_np_normalized = (img_np - img_mean) / img_std

# (H, W, C) -> (C, H, W)
img_np_transposed = img_np_normalized.transpose(2, 0, 1)

batch_img = [img_np_transposed]

ort_inputs = {ort_session.get_inputs()[0].name: batch_img}
ort_outs = ort_session.run(None, ort_inputs)[0]
batch_result = np.argmax(ort_outs, axis=1)
print(batch_result)

このSqueezenetモデルを使って昨日と同じようなことをするviews.pyが以下のようになります。全文はGitHubを見てください。 github.com

import os
import re
import urllib.request
from urllib.parse import urlparse
from PIL import Image
from joblib import dump, load
import tweepy

from django.shortcuts import render
from social_django.models import UserSocialAuth
from django.conf import settings
import more_itertools

import numpy as np
import onnxruntime
import torchvision.transforms as transforms


def to_numpy(tensor):
    return tensor.detach().cpu().numpy() if tensor.requires_grad else tensor.cpu().numpy()


ort_session = onnxruntime.InferenceSession(
    os.path.join(os.path.dirname(__file__), "model.onnx"))

data_transforms = transforms.Compose([
    transforms.RandomResizedCrop(224),
    transforms.ToTensor(),
    transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])


def index(request):
    if request.user.is_authenticated:
        user = UserSocialAuth.objects.get(user_id=request.user.id)
        consumer_key = settings.SOCIAL_AUTH_TWITTER_KEY
        consumer_secret = settings.SOCIAL_AUTH_TWITTER_SECRET
        access_token = user.extra_data['access_token']['oauth_token']
        access_secret = user.extra_data['access_token']['oauth_token_secret']
        auth = tweepy.OAuthHandler(consumer_key, consumer_secret)
        auth.set_access_token(access_token, access_secret)
        api = tweepy.API(auth)
        timeline = api.home_timeline(count=200, tweet_mode='extended')

        tweet_media = []
        for tweet in timeline:
            if 'media' in tweet.entities:
                tweet_media.append(tweet)

        batch_size = 4
        tweet_illust = []
        for batch_tweet in more_itertools.chunked(tweet_media, batch_size):
            batch_img = []
            for tweet in batch_tweet:
                media_url = tweet.extended_entities['media'][0]['media_url']
                filename = os.path.basename(urlparse(media_url).path)
                filename = os.path.join(
                    os.path.dirname(__file__), 'images', filename)
                urllib.request.urlretrieve(media_url, filename)
                img = Image.open(filename).convert('RGB')
                img = data_transforms(img)
                batch_img.append(to_numpy(img))

            ort_inputs = {ort_session.get_inputs()[0].name: batch_img}
            ort_outs = ort_session.run(None, ort_inputs)[0]
            batch_result = np.argmax(ort_outs, axis=1)
            for tweet, result in zip(batch_tweet, batch_result):
                if result == 1:
                    media_url = tweet.extended_entities['media'][0]['media_url']
                    if hasattr(tweet, "retweeted_status"):
                        profile_image_url = tweet.retweeted_status.author.profile_image_url_https
                        author = {'name': tweet.retweeted_status.author.name,
                                  'screen_name': tweet.retweeted_status.author.screen_name}
                        id_str = tweet.retweeted_status.id_str
                    else:
                        profile_image_url = tweet.author.profile_image_url_https
                        author = {'name': tweet.author.name,
                                  'screen_name': tweet.author.screen_name}
                        id_str = tweet.id_str
                    try:
                        text = tweet.retweeted_status.full_text
                    except AttributeError:
                        text = tweet.full_text
                    text = re.sub(
                        r"https?://[\w/:%#\$&\?\(\)~\.=\+\-]+$", '', text).rstrip()
                    tweet_illust.append({'id_str': id_str,
                                         'profile_image_url': profile_image_url,
                                         'author': author,
                                         'text': text,
                                         'image_url': media_url})
        tweet_illust_chunked = list(more_itertools.chunked(tweet_illust, 4))
        return render(request, 'hello/index.html', {'user': user, 'timeline_chunked': tweet_illust_chunked})
    else:
        return render(request, 'hello/index.html')

モデルがだいぶ小さくなったので、貧弱サーバーでも動かすことができました。

デプロイ

このモデルサイズならHerokuで動かせると思ったのですが、torchvisionなどの依存ライブラリの容量だけでHerokuの500MBの制限を超えてしまうようなので自分のサーバーで動かすことにしました。一応動いてはいるのですが、コールバックの設定がうまくいかないので後日直します。

2.9MBのモデルなら一瞬で推論できると思ったのですが、それでもまだ貧弱サーバーには荷が重いようで読み込みにだいぶ時間がかかります。もっと軽いモデルを作るかディープラーニングに頼らない方法を考えるのが良さそうです。

PyTorchのCPU版をrequirements.txtで指定すればHerokuにデプロイできました。

https://kivantium-playground.herokuapp.com/ から試すことができます。(開発状況によっては違うものがデプロイされているかもしれません)

広告コーナー