{
  "id": 293305,
  "title": "Configs - TensorFlow OD API",
  "url": "/competitions/tensorflow-great-barrier-reef/discussion/293305",
  "author_name": "",
  "post_date": "2021-12-05T02:53:03.514009100Z",
  "votes": 16,
  "comment_count": 3,
  "views": 0,
  "content": "<p>To make a stronger model with the object detection api, understanding config files is essential. <br>\n<a href=\"https://github.com/tensorflow/models/tree/master/research/object_detection/configs/tf2\" target=\"_blank\">Link to template model configs in tensorflow github</a></p>\n<p>To start, I would recommend using a config template from the model configs github (linked at the top) and changing it for the COTS use case (using steps below). Going through the configs in that link will also help you understand how they work.</p>\n<p><strong>There are 5 main parts to a config:</strong></p>\n<ul>\n<li>model</li>\n<li>train_config</li>\n<li>train_input_reader</li>\n<li>eval_config</li>\n<li>eval_input_reader</li>\n</ul>\n<p><strong>Model:</strong><br>\nIn this section you define what type of model you are using (e.g. faster_rcnn, ssd, center_net).<br>\nIn the feature_selction section you can set the type to whatever specific model you plan to use. For example, you can use something like ssd_efficientnet-b0_bifpn_keras or faster_rcnn_resnet50_keras.<br>\nThere is also a image resizer section where I set the min_dimension and max_dimension to 1280<br>\nDepending on your model, there are tons of other parameters you can change.</p>\n<p><strong>Train Config:</strong><br>\nIn this section it is essential to set the fine_tune_checkpoint and make sure the fine_tune_checkpoint_type is set to detection.<br>\nYou can also set some basic data augmentation options.<br>\nYou can also define an optimizer and number of training steps in this section</p>\n<p><strong>Train Input Reader:</strong><br>\nIn this section you need to set the label_map_path to your file path to your label_map.pbtxt file<br>\nYou also need to set the input path to the train area of your tf_records dataset. </p>\n<p><strong>Eval Config:</strong><br>\nThere isn’t much to do here besides maybe changing the batch size.</p>\n<p><strong>Eval Input Reader:</strong><br>\nIn this section you need to set the label_map_path to your file path to your label_map.pbtxt file<br>\nYou also need to set the input path to the eval area of your tf_records dataset. </p>\n<p>I hope this post helps you customize your tensorflow object detection models. Feel free to comment if you have any questions.</p>",
  "messages": [
    {
      "id": "1606458",
      "postDate": "12/05/2021 02:53:03",
      "content": "<p>To make a stronger model with the object detection api, understanding config files is essential. <br>\n<a href=\"https://github.com/tensorflow/models/tree/master/research/object_detection/configs/tf2\" target=\"_blank\">Link to template model configs in tensorflow github</a></p>\n<p>To start, I would recommend using a config template from the model configs github (linked at the top) and changing it for the COTS use case (using steps below). Going through the configs in that link will also help you understand how they work.</p>\n<p><strong>There are 5 main parts to a config:</strong></p>\n<ul>\n<li>model</li>\n<li>train_config</li>\n<li>train_input_reader</li>\n<li>eval_config</li>\n<li>eval_input_reader</li>\n</ul>\n<p><strong>Model:</strong><br>\nIn this section you define what type of model you are using (e.g. faster_rcnn, ssd, center_net).<br>\nIn the feature_selction section you can set the type to whatever specific model you plan to use. For example, you can use something like ssd_efficientnet-b0_bifpn_keras or faster_rcnn_resnet50_keras.<br>\nThere is also a image resizer section where I set the min_dimension and max_dimension to 1280<br>\nDepending on your model, there are tons of other parameters you can change.</p>\n<p><strong>Train Config:</strong><br>\nIn this section it is essential to set the fine_tune_checkpoint and make sure the fine_tune_checkpoint_type is set to detection.<br>\nYou can also set some basic data augmentation options.<br>\nYou can also define an optimizer and number of training steps in this section</p>\n<p><strong>Train Input Reader:</strong><br>\nIn this section you need to set the label_map_path to your file path to your label_map.pbtxt file<br>\nYou also need to set the input path to the train area of your tf_records dataset. </p>\n<p><strong>Eval Config:</strong><br>\nThere isn’t much to do here besides maybe changing the batch size.</p>\n<p><strong>Eval Input Reader:</strong><br>\nIn this section you need to set the label_map_path to your file path to your label_map.pbtxt file<br>\nYou also need to set the input path to the eval area of your tf_records dataset. </p>\n<p>I hope this post helps you customize your tensorflow object detection models. Feel free to comment if you have any questions.</p>",
      "rawMarkdown": "To make a stronger model with the object detection api, understanding config files is essential. \n[Link to template model configs in tensorflow github](https://github.com/tensorflow/models/tree/master/research/object_detection/configs/tf2)\n\nTo start, I would recommend using a config template from the model configs github (linked at the top) and changing it for the COTS use case (using steps below). Going through the configs in that link will also help you understand how they work.\n\n**There are 5 main parts to a config:**\n- model\n- train_config\n- train_input_reader\n- eval_config\n- eval_input_reader\n\n**Model:**\nIn this section you define what type of model you are using (e.g. faster_rcnn, ssd, center_net).\nIn the feature_selction section you can set the type to whatever specific model you plan to use. For example, you can use something like ssd_efficientnet-b0_bifpn_keras or faster_rcnn_resnet50_keras.\nThere is also a image resizer section where I set the min_dimension and max_dimension to 1280\nDepending on your model, there are tons of other parameters you can change.\n\n**Train Config:**\nIn this section it is essential to set the fine_tune_checkpoint and make sure the fine_tune_checkpoint_type is set to detection.\nYou can also set some basic data augmentation options.\nYou can also define an optimizer and number of training steps in this section\n\n**Train Input Reader:**\nIn this section you need to set the label_map_path to your file path to your label_map.pbtxt file\nYou also need to set the input path to the train area of your tf_records dataset. \n\n**Eval Config:**\nThere isn’t much to do here besides maybe changing the batch size.\n\n**Eval Input Reader:**\nIn this section you need to set the label_map_path to your file path to your label_map.pbtxt file\nYou also need to set the input path to the eval area of your tf_records dataset. \n\nI hope this post helps you customize your tensorflow object detection models. Feel free to comment if you have any questions.",
      "votes": null
    },
    {
      "id": "1612356",
      "postDate": "12/08/2021 19:49:16",
      "content": "<p>Thank you for this!</p>",
      "rawMarkdown": "Thank you for this!",
      "votes": null
    },
    {
      "id": "1620953",
      "postDate": "12/17/2021 09:29:36",
      "content": "<p>hi, tnx for ur notebook. It was very useful. Do you know any other resource to read more about configs?</p>",
      "rawMarkdown": "hi, tnx for ur notebook. It was very useful. Do you know any other resource to read more about configs?",
      "votes": null
    },
    {
      "id": "1621698",
      "postDate": "12/18/2021 01:07:37",
      "content": "<p><a href=\"https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/configuring_jobs.md\" target=\"_blank\">Here is the official documentation on config files</a><br>\n<a href=\"https://github.com/tensorflow/models/tree/master/research/object_detection/protos\" target=\"_blank\">Here are the proto files that tell you what the specific options are</a><br>\n<a href=\"https://towardsdatascience.com/3-steps-to-update-parameters-of-faster-r-cnn-ssd-models-in-tensorflow-object-detection-api-7eddb11273ed\" target=\"_blank\">Here is a more general toward data science article about them</a><br>\nHope this helps</p>",
      "rawMarkdown": "[Here is the official documentation on config files](https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/configuring_jobs.md)\n[Here are the proto files that tell you what the specific options are](https://github.com/tensorflow/models/tree/master/research/object_detection/protos)\n[Here is a more general toward data science article about them](https://towardsdatascience.com/3-steps-to-update-parameters-of-faster-r-cnn-ssd-models-in-tensorflow-object-detection-api-7eddb11273ed)\nHope this helps",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1612356,
      "author_name": "jacklinetum",
      "author_url": "",
      "post_date": "12/08/2021 19:49:16",
      "content": "<p>Thank you for this!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1620953,
      "author_name": "sarafadaei",
      "author_url": "",
      "post_date": "12/17/2021 09:29:36",
      "content": "<p>hi, tnx for ur notebook. It was very useful. Do you know any other resource to read more about configs?</p>",
      "votes": null,
      "replies": [
        {
          "id": 1621698,
          "author_name": "ravishah1",
          "author_url": "",
          "post_date": "12/18/2021 01:07:37",
          "content": "<p><a href=\"https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/configuring_jobs.md\" target=\"_blank\">Here is the official documentation on config files</a><br>\n<a href=\"https://github.com/tensorflow/models/tree/master/research/object_detection/protos\" target=\"_blank\">Here are the proto files that tell you what the specific options are</a><br>\n<a href=\"https://towardsdatascience.com/3-steps-to-update-parameters-of-faster-r-cnn-ssd-models-in-tensorflow-object-detection-api-7eddb11273ed\" target=\"_blank\">Here is a more general toward data science article about them</a><br>\nHope this helps</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1606458": "To make a stronger model with the object detection api, understanding config files is essential. \n[Link to template model configs in tensorflow github](https://github.com/tensorflow/models/tree/master/research/object_detection/configs/tf2)\n\nTo start, I would recommend using a config template from the model configs github (linked at the top) and changing it for the COTS use case (using steps below). Going through the configs in that link will also help you understand how they work.\n\n**There are 5 main parts to a config:**\n- model\n- train_config\n- train_input_reader\n- eval_config\n- eval_input_reader\n\n**Model:**\nIn this section you define what type of model you are using (e.g. faster_rcnn, ssd, center_net).\nIn the feature_selction section you can set the type to whatever specific model you plan to use. For example, you can use something like ssd_efficientnet-b0_bifpn_keras or faster_rcnn_resnet50_keras.\nThere is also a image resizer section where I set the min_dimension and max_dimension to 1280\nDepending on your model, there are tons of other parameters you can change.\n\n**Train Config:**\nIn this section it is essential to set the fine_tune_checkpoint and make sure the fine_tune_checkpoint_type is set to detection.\nYou can also set some basic data augmentation options.\nYou can also define an optimizer and number of training steps in this section\n\n**Train Input Reader:**\nIn this section you need to set the label_map_path to your file path to your label_map.pbtxt file\nYou also need to set the input path to the train area of your tf_records dataset. \n\n**Eval Config:**\nThere isn’t much to do here besides maybe changing the batch size.\n\n**Eval Input Reader:**\nIn this section you need to set the label_map_path to your file path to your label_map.pbtxt file\nYou also need to set the input path to the eval area of your tf_records dataset. \n\nI hope this post helps you customize your tensorflow object detection models. Feel free to comment if you have any questions.",
    "1612356": "Thank you for this!",
    "1620953": "hi, tnx for ur notebook. It was very useful. Do you know any other resource to read more about configs?",
    "1621698": "[Here is the official documentation on config files](https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/configuring_jobs.md)\n[Here are the proto files that tell you what the specific options are](https://github.com/tensorflow/models/tree/master/research/object_detection/protos)\n[Here is a more general toward data science article about them](https://towardsdatascience.com/3-steps-to-update-parameters-of-faster-r-cnn-ssd-models-in-tensorflow-object-detection-api-7eddb11273ed)\nHope this helps"
  },
  "source": "meta"
}