标签: object-detection

有没有办法在 Tensorflow 的对象检测 API 中使用针对 RGB 图像进行训练的预训练模型，用于单通道灰度图像（深度）？

我正在尝试阅读这一行中Tensorflow非极大值抑制方法的源代码。它是从gen_image_ops文件导入的，但我在张量流源代码中找不到该文件。

有什么来源可以让我获得这个方法的代码吗？

我一直在我自己的数据集上使用Tensorflow Object Detection API。在训练时，我想知道神经网络从训练集中学习的情况如何。因此，我想对训练集和评估集进行评估，并在训练期间分别获得准确率 (mAP)。

我的配置文件：

model {
  faster_rcnn {
    num_classes: 50
    image_resizer {
      fixed_shape_resizer {
        height: 960
        width: 960
      }
    }
    number_of_stages: 3
    feature_extractor {
      type: 'faster_rcnn_resnet101'
      first_stage_features_stride: 8
    }
    first_stage_anchor_generator {
      grid_anchor_generator {
        scales: [0.25, 0.5, 1.0, 2.0]
        aspect_ratios: [0.5, 1.0, 2.0]
        height_stride: 8
        width_stride: 8
      }
    }
    first_stage_atrous_rate: 2
    first_stage_box_predictor_conv_hyperparams {
      op: CONV
      regularizer {
        l2_regularizer {
          weight: 0.0
        }
      }
      initializer {
        truncated_normal_initializer {
          stddev: 0.00999999977648
        }
      }
    }
    first_stage_nms_score_threshold: 0.0
    first_stage_nms_iou_threshold: 0.699999988079
    first_stage_max_proposals: …

我正在自定义数据集（即车牌数据集）上训练张量流对象检测 API 模型。我的目标是使用 Tensorflow lite 将此模型部署到边缘设备，这样我就无法使用任何 RCNN 系列模型。因为，我无法将任何 RCNN 系列对象检测模型转换为 TensorFlow lite 模型（这是 TensorFlow 对象检测 API 的限制）。我正在使用ssd_mobilenet_v2_coco模型来训练自定义数据集。以下是我的配置文件的代码片段：

model {
  ssd {
    num_classes: 1
    box_coder {
      faster_rcnn_box_coder {
        y_scale: 10.0
        x_scale: 10.0
        height_scale: 5.0
        width_scale: 5.0
      }
    }
    matcher {
      argmax_matcher {
        matched_threshold: 0.5
        unmatched_threshold: 0.5
        ignore_thresholds: false
        negatives_lower_than_unmatched: true
        force_match_for_each_row: true
      }
    }
    similarity_calculator {
      iou_similarity {
      }
    }
    anchor_generator {
      ssd_anchor_generator {
        num_layers: 6
        min_scale: 0.2
        max_scale: 0.95
        aspect_ratios: 1.0
        aspect_ratios: 2.0
        aspect_ratios: 0.5
        aspect_ratios: …

我正在尝试按照此链接中的步骤安装 Tensorflow 对象检测 API ，这是 Tensorflow 2 的官方安装文档。

git clone https://github.com/tensorflow/models.git
> everything is ok
cd models/research/
> everything is ok
protoc object_detection/protos/*.proto --python_out=.
> everything is ok
cp object_detection/packages/tf2/setup.py .
> everything is ok
python -m pip install --use-feature=2020-resolver .

> Usage:   
>   /opt/anaconda3/envs/ml/bin/python -m pip install [options] <requirement specifier> [package-> index-options] ...
>   /opt/anaconda3/envs/ml/bin/python -m pip install [options] -r <requirements file> [package-index-options] ...
>   /opt/anaconda3/envs/ml/bin/python -m pip install [options] [-e] <vcs project url> ...
>   /opt/anaconda3/envs/ml/bin/python …

我正在尝试使用以下命令编译 protoc 文件：

protoc/bin/protoc models/research/object_detection/protos/*.proto --python_out=.

但我在 cmd 上得到这个输出

object_detection/protos/flexible_grid_anchor_generator.proto: File not found.
object_detection/protos/grid_anchor_generator.proto: File not found.
object_detection/protos/multiscale_anchor_generator.proto: File not found.
object_detection/protos/ssd_anchor_generator.proto: File not found.
models/research/object_detection/protos/anchor_generator.proto:5:1: Import "object_detection/protos/flexible_grid_anchor_generator.proto" was not found or had errors.
models/research/object_detection/protos/anchor_generator.proto:6:1: Import "object_detection/protos/grid_anchor_generator.proto" was not found or had errors.
models/research/object_detection/protos/anchor_generator.proto:7:1: Import "object_detection/protos/multiscale_anchor_generator.proto" was not found or had errors.
models/research/object_detection/protos/anchor_generator.proto:8:1: Import "object_detection/protos/ssd_anchor_generator.proto" was not found or had errors.
models/research/object_detection/protos/anchor_generator.proto:14:5: "GridAnchorGenerator" is not defined.
models/research/object_detection/protos/anchor_generator.proto:15:5: "SsdAnchorGenerator" is not defined.
models/research/object_detection/protos/anchor_generator.proto:16:5: "MultiscaleAnchorGenerator" is not defined.
models/research/object_detection/protos/anchor_generator.proto:17:5: "FlexibleGridAnchorGenerator" is not …

我有大约 50000 个图像和注释文件用于训练 YOLOv5 对象检测模型。我在另一台计算机上仅使用 CPU 训练模型没有问题，但需要太长时间，因此我需要 GPU 训练。我的问题是，当我尝试使用 GPU 进行训练时，我不断收到此错误：

OSError: [WinError 1455] The paging file is too small for this operation to complete

这是我正在执行的命令：

train.py --img 640 --batch 4 --epochs 100 --data myyaml.yaml --weights yolov5l.pt

CUDA和PyTorch已成功安装并可用。以下命令安装没有错误：

pip3 install torch==1.10.0+cu113 torchvision==0.11.1+cu113 torchaudio===0.10.0+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html

num_workers = 8我在网上发现其他人也有类似的问题，并通过将 更改为 来修复它num_workers = 1。当我尝试这个时，训练开始了，似乎已经过了出现错误的点the paging file is too small，但几个小时后崩溃了。我还根据此视频 ( https://www.youtube.com/watch?v=Oh6dga-Oy10 ) 增加了 GPU 上可用的虚拟内存，但这也不起作用。我认为这是一个内存问题，因为有时它崩溃时我会从计算机收到内存不足的警告。

任何帮助将非常感激。

我正在尝试使用 Pytorch 训练我自己的对象检测模型。但我总是遇到这个错误。我尝试更改火炬版本，但这没有帮助。
我的软件包：torchvision-0.11.1 和 torch-1.10.0

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-10-9e52b782b448> in <module>()
      4 for epoch in range(num_epochs):
      5     # training for one epoch
----> 6     train_one_epoch(model, optimizer, data_loader, device, epoch, print_freq=10)
      7     # update the learning rate
      8     lr_scheduler.step()

/content/engine.py in train_one_epoch(model, optimizer, data_loader, device, epoch, print_freq)
     21         warmup_iters = min(1000, len(data_loader) - 1)
     22 
---> 23         lr_scheduler = torch.optim.lr_scheduler.LinearLR(
     24             optimizer, start_factor=warmup_factor, total_iters=warmup_iters
     25         )

AttributeError: module 'torch.optim.lr_scheduler' has no attribute 'LinearLR'

我按照本教程进行对象检测： https://pytorch.org/tutorials/intermediate/torchvision_tutorial.html

及其 GitHub 存储库，其中包含以下train_one_epoch功能evaluate：

https://github.com/pytorch/vision/blob/main/references/detection/engine.py

但是，我想计算验证期间的损失。我为评估损失实现了这一点，本质上要获得损失，model.train()需要：

@torch.no_grad()
def evaluate_loss(model, data_loader, device):
    val_loss = 0
    model.train()
    for images, targets in data_loader:
        images = list(image.to(device) for image in images)
        targets = [{k: v.to(device) for k, v in t.items()} for t in targets]

        loss_dict = model(images, targets)

        losses = sum(loss for loss in loss_dict.values())

        # reduce losses over all GPUs for logging purposes
        loss_dict_reduced = utils.reduce_dict(loss_dict)
        losses_reduced = sum(loss for loss in loss_dict_reduced.values())
        val_loss += losses_reduced …

如何加载自定义 yolo v-7 模型。

这就是我知道如何加载 yolo v-5 模型的方法：

model = torch.hub.load('ultralytics/yolov5', 'custom', path='yolov5/runs/train/exp15/weights/last.pt', force_reload=True)

我在网上看到视频，他们建议使用这个：

!python detect.py --weights runs/train/yolov7x-custom/weights/best.pt --conf 0.5 --img-size 640 --source final_test_v1.mp4 

但我希望它像普通模型一样加载，并给我它找到对象的边界框坐标。

这就是我在 yolo v-5 中的做法：

from models.experimental import attempt_load
yolov5_weight_file = r'weights/rider_helmet_number_medium.pt' # ... may need full path
model = attempt_load(yolov5_weight_file, map_location=device)

def object_detection(frame):
    img = torch.from_numpy(frame)
    img = img.permute(2, 0, 1).float().to(device)  #convert to required shape based on index
    img /= 255.0  
    if img.ndimension() == 3:
        img = img.unsqueeze(0)

    pred = model(img, augment=False)[0]
    pred = …