FastRCNN 实现对 VOCdevkit 文件夹内好坏香蕉目标识别+bounding box主要代码

以下是使用 PyTorch 实现 Fast R-CNN 对 VOCdevkit 数据集中的香蕉目标进行识别和边界框定位的主要代码：

1. 准备数据集：

import torchvision
from torchvision.datasets import VOCDetection

# 数据集路径
data_dir = "/path/to/VOCdevkit/"

# 数据集类别
classes = ['bad_banana', 'good_banana']

# 加载 VOCdevkit 数据集
voc_dataset = VOCDetection(data_dir, year="2007", image_set="trainval", 
                           transform=torchvision.transforms.ToTensor())

# 定义数据加载器
batch_size = 4
data_loader = torch.utils.data.DataLoader(voc_dataset, batch_size=batch_size, 
                                          shuffle=True, num_workers=4)

2. 模型训练：

import torch
import torchvision.models as models
import torch.optim as optim
import torch.nn.functional as F
from torch.autograd import Variable
from torchvision.models.detection import fasterrcnn_resnet50_fpn

# 训练参数
learning_rate = 0.001
num_epochs = 10

# 加载预训练的 Faster R-CNN 模型
model = fasterrcnn_resnet50_fpn(pretrained=True)

# 替换分类器头部
num_classes = len(classes) + 1
in_features = model.roi_heads.box_predictor.cls_score.in_features
model.roi_heads.box_predictor = FastRCNNPredictor(in_features, num_classes)

# 定义优化器和损失函数
optimizer = optim.SGD(model.parameters(), lr=learning_rate, momentum=0.9)
criterion = torch.nn.CrossEntropyLoss()

# 训练模型
for epoch in range(num_epochs):
    running_loss = 0.0
    for i, data in enumerate(data_loader, 0):
        # 加载数据
        inputs, labels = data
        inputs, labels = Variable(inputs), Variable(labels)

        # 前向传播
        outputs = model(inputs)
        loss = criterion(outputs['loss_classifier'], labels)

        # 反向传播和优化
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        # 统计损失
        running_loss += loss.data[0]

    # 输出训练状态
    print('[%d, %5d] loss: %.3f' % (epoch + 1, i + 1, running_loss / len(data_loader)))

3. 目标检测：

import cv2
import numpy as np
from torchvision.ops import box_iou

# 加载测试图像
image_path = "/path/to/test/image.jpg"
image = cv2.imread(image_path)

# 图像预处理
image_tensor = torchvision.transforms.ToTensor()(image)
image_tensor = image_tensor.unsqueeze(0)

# 使用训练好的模型进行目标检测
model.eval()
with torch.no_grad():
    outputs = model(image_tensor)

# 获取检测结果
boxes = outputs[0]['boxes'].cpu().numpy()
scores = outputs[0]['scores'].cpu().numpy()
labels = outputs[0]['labels'].cpu().numpy()

# 筛选香蕉目标
banana_indices = np.where(labels == 1)[0]
banana_boxes = boxes[banana_indices]
banana_scores = scores[banana_indices]

# 使用 NMS 筛选较好的香蕉目标
nms_threshold = 0.5
keep_indices = torchvision.ops.nms(torch.from_numpy(banana_boxes), torch.from_numpy(banana_scores), nms_threshold)
good_banana