8

我目前正在尝试使用 GTSDB 数据集训练 Faster RCNN Inception V2 模型（使用 COCO 预训练）。我有 FullIJCNN 数据集，我将数据集分为训练、验证和测试三部分。最后，我分别创建了 3 个不同的 csv 文件，然后为train和validation创建了 TFRecord 文件。另一方面，我有一个代码块，可以读取每个图像的真实框坐标，并在图像上的交通标志周围绘制框。它也正确地写了类标签。这里有几个例子。同样，这些框不是由网络预测的。他们通过一个函数手动绘制。

画框 1

画框 2

然后我使用数据集文件夹中包含的 README 文件创建了一个标签文件，并将0 背景行添加到 labels.txt 的第一行以使其与我的代码一起使用（我认为这是一件愚蠢的事情），因为它是抛出索引错误。但是，我的 .pbtxt 文件中没有“背景”的键以使其从 1 开始。最后我配置了faster_rcnn_inception_v2_coco.config文件，更改num_classes: 90为num_classes: 43因为数据集有 43 个类，num_examples: 5000因为num_examples: 186我已经将数据集划分为186 个测试示例。num_steps: 200000照原样使用。最后，我通过运行开始了培训工作

python object_detection/model_main.py \
    --pipeline_config_path=${PIPELINE_CONFIG_PATH} \
    --model_dir=${MODEL_DIR} \
    --num_train_steps=50000 \
    --num_eval_steps=2000 \
    --alsologtostderr

命令，这是回溯（对不起代码块，我不知道如何专门添加日志）：

import matplotlib; matplotlib.use('Agg')  # pylint: disable=multiple-statements
WARNING:tensorflow:Estimator's model_fn (<function model_fn at 0x7fc4cd6a4938>) includes params argument, but params are not passed to Estimator.
WARNING:tensorflow:num_readers has been reduced to 1 to match input file shards.
WARNING:tensorflow:From /home/models/research/object_detection/core/box_predictor.py:407: calling reduce_mean (from tensorflow.python.ops.math_ops) with keep_dims is deprecated and will be removed in a future version.
Instructions for updating:
keep_dims is deprecated, use keepdims instead
WARNING:tensorflow:From /home/models/research/object_detection/meta_architectures/faster_rcnn_meta_arch.py:2037: get_or_create_global_step (from tensorflow.contrib.framework.python.ops.variables) is deprecated and will be removed in a future version.
Instructions for updating:
Please switch to tf.train.get_or_create_global_step
WARNING:tensorflow:From /home/models/research/object_detection/core/losses.py:317: softmax_cross_entropy_with_logits (from tensorflow.python.ops.nn_ops) is deprecated and will be removed in a future version.
Instructions for updating:

Future major versions of TensorFlow will allow gradients to flow
into the labels input on backprop by default.

See @{tf.nn.softmax_cross_entropy_with_logits_v2}.

/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/gradients_impl.py:100: UserWarning: Converting sparse IndexedSlices to a dense Tensor of unknown shape. This may consume a large amount of memory.
  "Converting sparse IndexedSlices to a dense Tensor of unknown shape. "
2018-07-26 09:48:21.785041: I tensorflow/core/platform/cpu_feature_guard.cc:140] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA
2018-07-26 09:48:21.923329: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1356] Found device 0 with properties:
name: Tesla K80 major: 3 minor: 7 memoryClockRate(GHz): 0.8235
pciBusID: 9b2f:00:00.0
totalMemory: 11.17GiB freeMemory: 11.10GiB
2018-07-26 09:48:21.923382: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1435] Adding visible gpu devices: 0
2018-07-26 09:48:22.153991: I tensorflow/core/common_runtime/gpu/gpu_device.cc:923] Device interconnect StreamExecutor with strength 1 edge matrix:
2018-07-26 09:48:22.154053: I tensorflow/core/common_runtime/gpu/gpu_device.cc:929]      0
2018-07-26 09:48:22.154075: I tensorflow/core/common_runtime/gpu/gpu_device.cc:942] 0:   N
2018-07-26 09:48:22.154333: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1053] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 10763 MB memory) -> physical GPU (device: 0, name: Tesla K80, pci bus id: 9b2f:00:00.0, compute capability: 3.7)
2018-07-26 09:58:31.794649: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1435] Adding visible gpu devices: 0
2018-07-26 09:58:31.794723: I tensorflow/core/common_runtime/gpu/gpu_device.cc:923] Device interconnect StreamExecutor with strength 1 edge matrix:
2018-07-26 09:58:31.794747: I tensorflow/core/common_runtime/gpu/gpu_device.cc:929]      0
2018-07-26 09:58:31.794765: I tensorflow/core/common_runtime/gpu/gpu_device.cc:942] 0:   N
2018-07-26 09:58:31.794884: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1053] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 10763 MB memory) -> physical GPU (device: 0, name: Tesla K80, pci bus id: 9b2f:00:00.0, compute capability: 3.7)
WARNING:tensorflow:Ignoring ground truth with image id 2066941970 since it was previously added
WARNING:tensorflow:Ignoring detection with image id 2066941970 since it was previously added
WARNING:tensorflow:Ignoring ground truth with image id 2013299735 since it was previously added
WARNING:tensorflow:Ignoring detection with image id 2013299735 since it was previously added
WARNING:tensorflow:Ignoring ground truth with image id 1416415107 since it was previously added

它产生了很多这样的警告：

WARNING:tensorflow:Ignoring ground truth with image id 2013299735 since it was previously added
WARNING:tensorflow:Ignoring detection with image id 2013299735 since it was previously added

这些消息的原因num_examples已设置为2000尽管我的原始配置文件有num_examples: 186. 我不明白为什么它要创建一个具有不同参数的新配置文件。然而，在充满这些消息的整个日志之后，它会给出一个报告，但我不能确定这到底是想告诉我什么。这是报告：

creating index...
index created!
creating index...
index created!
Running per image evaluation...
Evaluate annotation type *bbox*
DONE (t=0.07s).
Accumulating evaluation results...
DONE (t=0.02s).
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.000
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.000
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.000
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.000
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = -1.000
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = -1.000
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.000
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.000
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.000
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.000
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = -1.000
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = -1.000

最后，我检查了 Tensorboard 以确保它训练正确，但我看到的结果令人沮丧。这是我的模型（损失）的 Tensorboard 图的屏幕截图：

失利

一般损失

我觉得我做错了什么。我不知道这是否是一个具体的问题，但我试图尽可能详细地提供。

我的问题是：我应该在这些步骤中进行哪些更改？为什么我的函数绘制了真正的盒子，但我的模型无法弄清楚发生了什么？提前致谢！

4

4 回答 4

10

您收到警告的原因是因为您的数据集中的项目正在被多次评估。您为 num_train_steps 和 num_eval_steps 指定的值应与您的 train_config batch_size 和数据集的大小相关。例如，如果您的批量大小为 24，并且您有 24000 条训练记录，则 num_train_steps 应设置为 1000，同样的 num_eval_steps 计算方法相同，但具有评估记录的数量。如果您使用指定的值执行脚本，model_main.py 脚本似乎没有利用您在 pipeline.config 文件中指定的值。

于 2018-07-26T16:17:36.543 回答
4

我遇到了同样的问题，过了一会儿，我想出了这个对我有用但不能成为全局解决方案的解决方案；如果您使用的是分布在多个文件夹中的数据集，并且您使用的是您自己制作的 tf_record 转换器，则可能会在整个数据集中的每个帧命名冲突中出现问题。

由于我使用完整路径作为文件名（避免冲突），我再也没有看到警告。我希望它会帮助某人。

tf_example = tf.train.Example(features=tf.train.Features(feature={
    'image/height': dataset_util.int64_feature(im_height),
    'image/width': dataset_util.int64_feature(im_width),
    'image/filename': dataset_util.bytes_feature(filename),
    'image/source_id': dataset_util.bytes_feature(filename),
    'image/encoded': dataset_util.bytes_feature(encoded_image_data),
    'image/format': dataset_util.bytes_feature(image_format),
    'image/object/bbox/xmin': dataset_util.float_list_feature(xmins),
    'image/object/bbox/xmax': dataset_util.float_list_feature(xmaxs),
    'image/object/bbox/ymin': dataset_util.float_list_feature(ymins),
    'image/object/bbox/ymax': dataset_util.float_list_feature(ymaxs),
    'image/object/class/text': dataset_util.bytes_list_feature(classes_text),
    'image/object/class/label': dataset_util.int64_list_feature(classes),
}))
于 2019-02-23T00:48:02.050 回答
0

如果警告仍然存在，请再次检查您的 generate_tfrecord.py 文件。为了方便起见，我更改了 tfrecord 文件，其中有一些错误。我建议只检查您的 tfrecord 文件是否正确显示了基本事实框。然后只进行培训。

于 2019-06-06T05:34:32.367 回答
0

我确实遇到了示例问题，是的，在我更改num_example以匹配我的验证案例后，错误就消失了。

于 2019-07-19T15:46:18.480 回答