为啥 TensorFlow 对象检测 2.x 在训练模型时不显示 mAP

Posted

技术标签:

【中文标题】为啥 TensorFlow 对象检测 2.x 在训练模型时不显示 mAP【英文标题】:Why TensorFlow object detection 2.x don't show mAP when training the model为什么 TensorFlow 对象检测 2.x 在训练模型时不显示 mAP 【发布时间】:2021-09-26 00:37:49 【问题描述】:

我以前用TF 1.4训练过一些物体检测模型,我记得训练时的评估显示了模型的mAP。我的问题是,现在,在 TF 2.5 上,这些指标没有显示,我需要它来评估我的成功。这是我唯一的输出:

I0715 00:57:35.858141 140071375349632 model_lib_v2.py:701] 'Loss/classification_loss': 0.19326138,
 'Loss/localization_loss': 0.07984769,
 'Loss/regularization_loss': 0.2631261,
 'Loss/total_loss': 0.5362352,
 'learning_rate': 0.03066655

我已经对模型进行了 2k 步的训练,但什么也没有……我无法仅根据损失来评估我的模型。如何再次打印 mAP?

这是我的管道配置文件(我使用的是带有 Resnet 50 的 SSD):

model 
  ssd 
    num_classes: 3
    image_resizer 
      fixed_shape_resizer 
        height: 640
        width: 640
      
    
    feature_extractor 
      type: "ssd_resnet50_v1_fpn_keras"
      depth_multiplier: 1.0
      min_depth: 16
      conv_hyperparams 
        regularizer 
          l2_regularizer 
            weight: 0.00039999998989515007
          
        
        initializer 
          truncated_normal_initializer 
            mean: 0.0
            stddev: 0.029999999329447746
          
        
        activation: RELU_6
        batch_norm 
          decay: 0.996999979019165
          scale: true
          epsilon: 0.0010000000474974513
        
      
      override_base_feature_extractor_hyperparams: true
      fpn 
        min_level: 3
        max_level: 7
      
    
    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
        use_matmul_gather: true
      
    
    similarity_calculator 
      iou_similarity 
      
    
    box_predictor 
      weight_shared_convolutional_box_predictor 
        conv_hyperparams 
          regularizer 
            l2_regularizer 
              weight: 0.00039999998989515007
            
          
          initializer 
            random_normal_initializer 
              mean: 0.0
              stddev: 0.009999999776482582
            
          
          activation: RELU_6
          batch_norm 
            decay: 0.996999979019165
            scale: true
            epsilon: 0.0010000000474974513
          
        
        depth: 256
        num_layers_before_predictor: 4
        kernel_size: 3
        class_prediction_bias_init: -4.599999904632568
      
    
    anchor_generator 
      multiscale_anchor_generator 
        min_level: 3
        max_level: 7
        anchor_scale: 4.0
        aspect_ratios: 1.0
        aspect_ratios: 2.0
        aspect_ratios: 0.5
        scales_per_octave: 2
      
    
    post_processing 
      batch_non_max_suppression 
        score_threshold: 9.99999993922529e-09
        iou_threshold: 0.6000000238418579
        max_detections_per_class: 100
        max_total_detections: 100
        use_static_shapes: false
      
      score_converter: SIGMOID
    
    normalize_loss_by_num_matches: true
    loss 
      localization_loss 
        weighted_smooth_l1 
        
      
      classification_loss 
        weighted_sigmoid_focal 
          gamma: 2.0
          alpha: 0.25
        
      
      classification_weight: 1.0
      localization_weight: 1.0
    
    encode_background_as_zeros: true
    normalize_loc_loss_by_codesize: true
    inplace_batchnorm_update: true
    freeze_batchnorm: false
  

train_config 
  batch_size: 8
  data_augmentation_options 
    random_horizontal_flip 
    
  
  data_augmentation_options 
    random_crop_image 
      min_object_covered: 0.0
      min_aspect_ratio: 0.75
      max_aspect_ratio: 3.0
      min_area: 0.75
      max_area: 1.0
      overlap_thresh: 0.0
    
  
  sync_replicas: true
  optimizer 
    momentum_optimizer 
      learning_rate 
        cosine_decay_learning_rate 
          learning_rate_base: 0.03999999910593033
          total_steps: 25000
          warmup_learning_rate: 0.013333000242710114
          warmup_steps: 2000
        
      
      momentum_optimizer_value: 0.8999999761581421
    
    use_moving_average: false
  
  fine_tune_checkpoint: "/content/models/research/pretrained_model/ssd_resnet50_v1_fpn_640x640_coco17_tpu-8/checkpoint/ckpt-0"
  num_steps: 2100
  startup_delay_steps: 0.0
  replicas_to_aggregate: 8
  max_number_of_boxes: 100
  unpad_groundtruth_tensors: false
  fine_tune_checkpoint_type: "detection"
  use_bfloat16: true
  fine_tune_checkpoint_version: V2

train_input_reader 
  label_map_path: "/content/label_map.pbtxt"
  tf_record_input_reader 
    input_path: "/content/train.record"
  

eval_config 
  metrics_set: "coco_detection_metrics"
  use_moving_averages: false

eval_input_reader 
  label_map_path: "/content/label_map.pbtxt"
  shuffle: false
  num_epochs: 1
  tf_record_input_reader 
    input_path: "/content/test.record"
  

【问题讨论】:

【参考方案1】:

您需要同时在两个 shell 中运行 model_main_tf2.py 脚本。

在第一个 shell 中,使用参数 --model_dir--pipeline_config_path 运行它以进行训练,如下所示:

python model_main_tf2.py --model_dir my-model --pipeline_config_path my-model/pipeline.config --alsologtostderr

在第二个 shell 中,您需要传递一个名为 --checkpoint_dir 的额外参数,指向存储检查点的文件夹,如下所示:

python model_main_tf2.py --model_dir my-model --pipeline_config_path my-model/pipeline.config --checkpont_dir my-model

这将触发脚本的评估模式,TensorBoard 将开始显示 mAP 和召回指标。

【讨论】:

谢谢,毕竟这是我在关注的解决方案,我需要同时使用两个Colab notebook,不太实用,但是很管用!【参考方案2】:

在 TF 2.5 中,您可以使用model.summary to see model configuration . metrics (loss ,accuracy ,learning rate ) can be changed in model.compile 。您可以在 model.fit 操作期间实时查看参数值。附上以下文件供您参考 https://www.tensorflow.org/js/guide/models_and_layers https://www.tensorflow.org/guide/keras/train_and_evaluate ,您还可以在训练模型的同时根据默认指标创建自定义指标来测试模型

【讨论】:

以上是关于为啥 TensorFlow 对象检测 2.x 在训练模型时不显示 mAP的主要内容,如果未能解决你的问题,请参考以下文章

Tensorflow对象检测:为啥使用ssd mobilnet v1时图像中的位置会影响检测精度?

为啥来自 Tensorflow 2 对象检测 API 的微调模型上的 mAP 较低?

基于 Tensorflow 2.x 从零训练 15 点人脸关键点检测模型

基于 Tensorflow 2.x 从零训练 15 点人脸关键点检测模型

基于 Tensorflow 2.x 从零训练 15 点人脸关键点检测模型

我如何使用 tensorflow 对象检测来仅检测人员?