PyTorch專欄(八):微調基於torchvision 0.3的目標檢測模型

作者 | News

編輯 | 安可

出品 | 磐創AI團隊出品

【磐創AI 導讀】：本篇文章講解了PyTorch專欄的第四章中的微調基於torchvision 0.3的目標檢測模型。查看專欄歷史文章，請點擊下方藍色字體進入相應連結閱讀。查看關於本專欄的介紹：PyTorch專欄開篇。想要更多電子雜誌的機器學習，深度學習資源，大家歡迎點擊上方藍字關注我們的公眾號：磐創AI。

微調基於torchvision 0.3的目標檢測模型使用Sequence2Sequence網絡和注意力進行翻譯在這篇文章中，我們將微調在 Penn-Fudan 資料庫中對行人檢測和分割的已預先訓練的 Mask R-CNN 模型。它包含170個圖像和345個行人實例，我們 將用它來說明如何在 torchvision 中使用新功能，以便在自定義數據集上訓練實例分割模型。1.定義數據集對於訓練對象檢測的引用腳本，實例分割和人員關鍵點檢測要求能夠輕鬆支持添加新的自定義數據。數據集應該從標準的類torch.utils.data.Dataset 繼承而來，並實現_len和_getitem_我們要求的唯一特性是數據集的__getitem__應該返回：* 圖像：PIL圖像大小(H,W) * 目標：包含以下欄位的字典
<1> boxes(FloatTensor[N,4])：N邊框（bounding boxes）坐標的格式[x0,x1,y0,y1]，取值範圍是0到W,0到H。
<2> labels(Int64Tensor[N])：每個邊框的標籤。
<3> image_id(Int64Tensor[1])：圖像識別器，它應該在數據集中的所有圖像中是唯一的，並在評估期間使用。
<4> area(Tensor[N])：邊框的面積，在使用COCO指標進行評估時使用此項來分隔小、中和大框之間的度量標準得分。
<5> iscrowed(UInt8Tensor[N,H,W])：在評估期間屬性設置為iscrowed=True的實例會被忽略。
<6> (可選)masks(UInt8Tesor[N,H,W])：每個對象的分段掩碼。
<7> (可選)keypoints (FloatTensor[N, K, 3]：對於N個對象中的每一個，它包含[x，y，visibility]格式的K個關鍵點，用 於定義對象。visibility = 0表示關鍵點不可見。請注意，對於數據擴充，翻轉關鍵點的概念取決於數據表示，您應該調整 reference/detection/transforms.py 以用於新的關鍵點表示。
如果你的模型返回上述方法，它們將使其適用於培訓和評估，並將使用 pycocotools 的評估腳本。此外，如果要在訓練期間使用寬高比分組（以便每個批次僅包含具有相似寬高比的圖像），則建議還實現get_height_and_width方法， 該方法返回圖像的高度和寬度。如果未提供此方法，我們將通過__getitem__查詢數據集的所有元素，這會將圖像加載到內存中，但比提供自定義方法時要慢。2.為 PennFudan 編寫自定義數據集2.1 下載數據集

PennFudanPed/
PedMasks/
FudanPed00001_mask.png
FudanPed00002_mask.png
FudanPed00003_mask.png
FudanPed00004_mask.png
...
PNGImages/
FudanPed00001.png
FudanPed00002.png
FudanPed00003.png
FudanPed00004.png

下面是一個圖像以及其分割掩膜的例子：  因此每個圖像具有相應的分割掩膜，其中每個顏色對應於不同的實例。讓我們為這個數據集寫一個torch.utils.data.Dataset類。2.2 為數據集編寫類

import os
import numpy as np
import torch
from PIL import Image


class PennFudanDataset(object):
    def __init__(self, root, transforms):
        self.root = root
        self.transforms = transforms
        
        
        self.imgs = list(sorted(os.listdir(os.path.join(root, "PNGImages"))))
        self.masks = list(sorted(os.listdir(os.path.join(root, "PedMasks"))))

    def __getitem__(self, idx):
        
        img_path = os.path.join(self.root, "PNGImages", self.imgs[idx])
        mask_path = os.path.join(self.root, "PedMasks", self.masks[idx])
        img = Image.open(img_path).convert("RGB")
        
        
        
        mask = Image.open(mask_path)
        
        mask = np.array(mask)
        
        obj_ids = np.unique(mask)
        
        obj_ids = obj_ids[1:]

        
        
        masks = mask == obj_ids[:, None, None]

        
        num_objs = len(obj_ids)
        boxes = []
        for i in range(num_objs):
            pos = np.where(masks[i])
            xmin = np.min(pos[1])
            xmax = np.max(pos[1])
            ymin = np.min(pos[0])
            ymax = np.max(pos[0])
            boxes.append([xmin, ymin, xmax, ymax])

        
        boxes = torch.as_tensor(boxes, dtype=torch.float32)
        
        labels = torch.ones((num_objs,), dtype=torch.int64)
        masks = torch.as_tensor(masks, dtype=torch.uint8)

        image_id = torch.tensor([idx])
        area = (boxes[:, 3] - boxes[:, 1]) * (boxes[:, 2] - boxes[:, 0])
        
        iscrowd = torch.zeros((num_objs,), dtype=torch.int64)

        target = {}
        target["boxes"] = boxes
        target["labels"] = labels
        target["masks"] = masks
        target["image_id"] = image_id
        target["area"] = area
        target["iscrowd"] = iscrowd

        if self.transforms is not None:
            img, target = self.transforms(img, target)

        return img, target

    def __len__(self):
        return len(self.imgs)

3.定義模型現在我們需要定義一個可以上述數據集執行預測的模型。在本教程中，我們將使用 Mask R-CNN， 它基於 Faster R-CNN。Faster R-CNN 是一種模型，可以預測圖像中潛在對象的邊界框和類別得分。 Mask R-CNN 在 Faster R-CNN 中添加了一個額外的分支，它還預測每個實例的分割蒙版。有兩種常見情況可能需要修改torchvision modelzoo中的一個可用模型。第一個是我們想要從預先訓練的模型開始，然後微調最後一層。另一種是當我們想要用不同的模型替換模型的主幹時（例如，用於更快的預測）。
3.1 PennFudan 數據集的實例分割模型在我們的例子中，我們希望從預先訓練的模型中進行微調，因為我們的數據集非常小，所以我們將遵循上述第一種情況。這裡我們還要計算實例分割掩膜，因此我們將使用 Mask R-CNN：

import torchvision
from torchvision.models.detection.faster_rcnn import FastRCNNPredictor
from torchvision.models.detection.mask_rcnn import MaskRCNNPredictor


def get_model_instance_segmentation(num_classes):
    
    model = torchvision.models.detection.maskrcnn_resnet50_fpn(pretrained=True)

    
    in_features = model.roi_heads.box_predictor.cls_score.in_features
    
    model.roi_heads.box_predictor = FastRCNNPredictor(in_features, num_classes)

    
    in_features_mask = model.roi_heads.mask_predictor.conv5_mask.in_channels
    hidden_layer = 256
    
    model.roi_heads.mask_predictor = MaskRCNNPredictor(in_features_mask,
                                                       hidden_layer,
                                                       num_classes)

    return model

就是這樣，這將使模型準備好在您的自定義數據集上進行訓練和評估。
4.整合在references/detection/中，我們有許多輔助函數來簡化訓練和評估檢測模型。在這裡，我們將使用 references/detection/engine.py，references/detection/utils.py和references/detection/transforms.py。只需將它們複製到您的文件夾並在此處使用它們。4.1 為數據擴充/轉換編寫輔助函數：

import transforms as T

def get_transform(train):
    transforms = []
    transforms.append(T.ToTensor())
    if train:
        transforms.append(T.RandomHorizontalFlip(0.5))
    return T.Compose(transforms)

4.2 編寫執行訓練和驗證的主要功能

from engine import train_one_epoch, evaluate
import utils


def main():
    
    device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')

    
    num_classes = 2
    
    dataset = PennFudanDataset('PennFudanPed', get_transform(train=True))
    dataset_test = PennFudanDataset('PennFudanPed', get_transform(train=False))

    
    indices = torch.randperm(len(dataset)).tolist()
    dataset = torch.utils.data.Subset(dataset, indices[:-50])
    dataset_test = torch.utils.data.Subset(dataset_test, indices[-50:])

    
    data_loader = torch.utils.data.DataLoader(
        dataset, batch_size=2, shuffle=True, num_workers=4,
        collate_fn=utils.collate_fn)

    data_loader_test = torch.utils.data.DataLoader(
        dataset_test, batch_size=1, shuffle=False, num_workers=4,
        collate_fn=utils.collate_fn)

    
    model = get_model_instance_segmentation(num_classes)

    
    model.to(device)

    
    params = [p for p in model.parameters() if p.requires_grad]
    optimizer = torch.optim.SGD(params, lr=0.005,
                                momentum=0.9, weight_decay=0.0005)
    
    lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer,
                                                   step_size=3,
                                                   gamma=0.1)

    
    num_epochs = 10

    for epoch in range(num_epochs):
        
        train_one_epoch(model, optimizer, data_loader, device, epoch, print_freq=10)
        
        lr_scheduler.step()
        
        evaluate(model, data_loader_test, device=device)

    print("That's it!")

Epoch: [0] [ 0/60] eta: 0:01:18 lr: 0.000090 loss: 2.5213 (2.5213) loss_classifier: 0.8025 (0.8025) loss_box_reg: 0.2634 (0.2634) loss_mask: 1.4265 (1.4265) loss_objectness: 0.0190 (0.0190) loss_rpn_box_reg: 0.0099 (0.0099) time: 1.3121 data: 0.3024 max mem: 3485
Epoch: [0] [10/60] eta: 0:00:20 lr: 0.000936 loss: 1.3007 (1.5313) loss_classifier: 0.3979 (0.4719) loss_box_reg: 0.2454 (0.2272) loss_mask: 0.6089 (0.7953) loss_objectness: 0.0197 (0.0228) loss_rpn_box_reg: 0.0121 (0.0141) time: 0.4198 data: 0.0298 max mem: 5081
Epoch: [0] [20/60] eta: 0:00:15 lr: 0.001783 loss: 0.7567 (1.1056) loss_classifier: 0.2221 (0.3319) loss_box_reg: 0.2002 (0.2106) loss_mask: 0.2904 (0.5332) loss_objectness: 0.0146 (0.0176) loss_rpn_box_reg: 0.0094 (0.0123) time: 0.3293 data: 0.0035 max mem: 5081
Epoch: [0] [30/60] eta: 0:00:11 lr: 0.002629 loss: 0.4705 (0.8935) loss_classifier: 0.0991 (0.2517) loss_box_reg: 0.1578 (0.1957) loss_mask: 0.1970 (0.4204) loss_objectness: 0.0061 (0.0140) loss_rpn_box_reg: 0.0075 (0.0118) time: 0.3403 data: 0.0044 max mem: 5081
Epoch: [0] [40/60] eta: 0:00:07 lr: 0.003476 loss: 0.3901 (0.7568) loss_classifier: 0.0648 (0.2022) loss_box_reg: 0.1207 (0.1736) loss_mask: 0.1705 (0.3585) loss_objectness: 0.0018 (0.0113) loss_rpn_box_reg: 0.0075 (0.0112) time: 0.3407 data: 0.0044 max mem: 5081
Epoch: [0] [50/60] eta: 0:00:03 lr: 0.004323 loss: 0.3237 (0.6703) loss_classifier: 0.0474 (0.1731) loss_box_reg: 0.1109 (0.1561) loss_mask: 0.1658 (0.3201) loss_objectness: 0.0015 (0.0093) loss_rpn_box_reg: 0.0093 (0.0116) time: 0.3379 data: 0.0043 max mem: 5081
Epoch: [0] [59/60] eta: 0:00:00 lr: 0.005000 loss: 0.2540 (0.6082) loss_classifier: 0.0309 (0.1526) loss_box_reg: 0.0463 (0.1405) loss_mask: 0.1568 (0.2945) loss_objectness: 0.0012 (0.0083) loss_rpn_box_reg: 0.0093 (0.0123) time: 0.3489 data: 0.0042 max mem: 5081
Epoch: [0] Total time: 0:00:21 (0.3570 s / it)
creating index...
index created!
Test: [ 0/50] eta: 0:00:19 model_time: 0.2152 (0.2152) evaluator_time: 0.0133 (0.0133) time: 0.4000 data: 0.1701 max mem: 5081
Test: [49/50] eta: 0:00:00 model_time: 0.0628 (0.0687) evaluator_time: 0.0039 (0.0064) time: 0.0735 data: 0.0022 max mem: 5081
Test: Total time: 0:00:04 (0.0828 s / it)
Averaged stats: model_time: 0.0628 (0.0687) evaluator_time: 0.0039 (0.0064)
Accumulating evaluation results...
DONE (t=0.01s).
Accumulating evaluation results...
DONE (t=0.01s).
IoU metric: bbox
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.606
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.984
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.780
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.313
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.582
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.612
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.270
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.672
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.672
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.650
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.755
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.664
IoU metric: segm
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.704
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.979
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.871
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.325
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.488
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.727
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.316
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.748
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.749
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.650
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.673
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.758

因此，在一個epoch訓練之後，我們獲得了COCO-style mAP為60.6，並且mask mAP為70.4。

IoU metric: bbox
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.799
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.969
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.935
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.349
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.592
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.831
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.324
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.844
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.844
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.400
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.777
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.870
IoU metric: segm
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.761
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.969
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.919
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.341
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.464
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.788
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.303
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.799
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.799
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.400
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.769
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.818

但預測結果如何呢？讓我們在數據集中拍攝一張圖像並進行驗證。 訓練的模型預測了此圖像中的9個人物，讓我們看看其中的幾個，由下圖可以看到預測效果很好。 5.總結在本教程中，您學習了如何在自定義數據集上為實例分段模型創建自己的訓練管道。為此，您編寫了一個torch.utils.data.Dataset類， 它返回圖像以及地面實況框和分割掩碼。您還利用了在COCO train2017上預訓練的Mask R-CNN模型，以便對此新數據集執行傳輸學習。有關包含multi-machine / multi-gpu training的更完整示例，請檢查 torchvision 存儲庫中的references/detection/train.py。 下方點擊 |  閱讀原文 | 了解更多