Tensorflow 对象检测 api 验证数据大小 [英] Tensorflow object detection api validation data size

问题描述

我正在从对象检测 API 运行教程，我正在使用带有 ResNet Faster-RCNN 的牛津数据集.

I am running tutorial from object detection API and I am using Oxford dataset with ResNet Faster-RCNN.

当我通过运行 (eval.py) 评估我的训练模型时，Tensorboard 返回大约 0.95 平滑精度值.

When I evaluate my trained model by running (eval.py), Tensorboard returns about 0.95 smoothed precision value.

我的问题是它评估了多少个图像集?因为来自 Tensorboard 及其教程链接(https://github.com/tensorflow/models/blob/master/object_detection/g3doc/running_pets.md)，Tensorboard 只显示 10 张图片.

My question is how many image set does it evaluate? Because from Tensorboard and their tutorial link (https://github.com/tensorflow/models/blob/master/object_detection/g3doc/running_pets.md), Tensorboard only shows 10 images.

这是否意味着他们只用 10 张图像检查精度?

Does it mean that they check precision only with 10 images?

我的牛津数据集验证 jpg 计数应该是大约 2,200.

My Oxford dataset validation jpg counts should be about 2,200.

在我的配置中，我像这样正确指定了输入路径:

In my configuration, I specified input path correctly like this:

eval_input_reader: {
  tf_record_input_reader {
    input_path: "my_path/pet_val.record"
  }
  label_map_path: "my_path/pet_label_map.pbtxt"
  shuffle: false
  num_readers: 1
}

eval.py 最后会打印 mAP 吗?

And does eval.py prints mAP at the end?

大约三天前我在我的 1 个 GPU 本地机器上运行 eval.py，但它没有打印任何内容.

I run eval.py about three days ago on my 1 GPU local machine, but it does not print anything.

最后，这个 API 是否提供 F 值和 fps(每秒帧数)?有人有这方面的经验吗?

Finally, does this API provide F-value and fps (frame per second)? Anyone has experience with this?

似乎我们可以从配置中设置评估大小限制，例如/object_detection/samples/configs/faster_rcnn_resnet101_pets.config#L131.当我从 https://github 打印 len(result_lists) 时.com/tensorflow/models/blob/master/object_detection/eval_util.py#L404，它打印 2000，这是我的 eval num_examples.

edit: it seems that we can setup eval size limit from configuration such as /object_detection/samples/configs/faster_rcnn_resnet101_pets.config#L131. When I print len(result_lists) from https://github.com/tensorflow/models/blob/master/object_detection/eval_util.py#L404, it prints 2000, which was my eval num_examples.

我还能够通过手动比较时间戳来生成 fps.

I was also able to generate fps by comparing timestamp manually.

推荐答案

默认情况下，我们只在 Tensorboard 上可视化 10 个图像(以避免被图像淹没)但这可以从 eval_config.您还可以在配置中更改评估的图像数量(默认为 5000).

By default, we only visualize 10 images on Tensorboard (to avoid overwhelming it with images) but this is configurable from the eval_config. You can also change the number of images evaluated (defaults to 5000) in the config too.

这篇关于Tensorflow 对象检测 api 验证数据大小的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！