如何理解 tensorflowjs 对象检测张量输出？

Question

我的动机是构建一个自定义的异议检测 Web 应用程序。我从model zoo 下载了tf2 预训练的SSD Resnet1010 模型。我的想法是，如果这个实现有效，我将用我自己的数据训练模型。我运行$saved_model_cli show --dir saved_model --tag_set serve --signature_def serving_default 来计算输入和输出节点。

The given SavedModel SignatureDef contains the following input(s):
  inputs['input_tensor'] tensor_info:
      dtype: DT_UINT8
      shape: (1, -1, -1, 3)
      name: serving_default_input_tensor:0
The given SavedModel SignatureDef contains the following output(s):
  outputs['detection_anchor_indices'] tensor_info:
      dtype: DT_FLOAT
      shape: (1, 100)
      name: StatefulPartitionedCall:0
  outputs['detection_boxes'] tensor_info:
      dtype: DT_FLOAT
      shape: (1, 100, 4)
      name: StatefulPartitionedCall:1
  outputs['detection_classes'] tensor_info:
      dtype: DT_FLOAT
      shape: (1, 100)
      name: StatefulPartitionedCall:2
  outputs['detection_multiclass_scores'] tensor_info:
      dtype: DT_FLOAT
      shape: (1, 100, 91)
      name: StatefulPartitionedCall:3
  outputs['detection_scores'] tensor_info:
      dtype: DT_FLOAT
      shape: (1, 100)
      name: StatefulPartitionedCall:4
  outputs['num_detections'] tensor_info:
      dtype: DT_FLOAT
      shape: (1)
      name: StatefulPartitionedCall:5
  outputs['raw_detection_boxes'] tensor_info:
      dtype: DT_FLOAT
      shape: (1, 51150, 4)
      name: StatefulPartitionedCall:6
  outputs['raw_detection_scores'] tensor_info:
      dtype: DT_FLOAT
      shape: (1, 51150, 91)
      name: StatefulPartitionedCall:7
Method name is: tensorflow/serving/predict

然后我通过运行将模型转换为 tensorflowjs 模型

tensorflowjs_converter --input_format=tf_saved_model --output_node_names='detection_anchor_indices,detection_boxes,detection_classes,detection_multiclass_scores,detection_scores,num_detections,raw_detection_boxes,raw_detection_scores' --saved_model_tags=serve --output_format=tfjs_graph_model saved_model js_model

这是我的 javascript 代码（在 vue 方法中）

    loadTfModel: async function(){
        try {
            this.model = await tf.loadGraphModel(this.MODEL_URL);
        } catch(error) {
            console.log(error);
        }

   },
    predictImg: async function() {
        const imgData = document.getElementById('img');
        let tf_img = tf.browser.fromPixels(imgData);
        tf_img = tf_img.expandDims(0);
        const predictions = await this.model.executeAsync(tf_img);
        const data = []
        for (let i = 0; i < predictions.length; i++){
            data.push(predictions[i].dataSync());
        }
        console.log(data);
    }

输出如下所示：

我的问题是数组中的这八个项目是否对应八个定义的输出节点？如何理解这些数据？以及如何将其转换为像 python 那样的人类可读格式？
更新 1： 我试过这个answer 并编辑了我的预测方法：

predictImg: async function() {
        const imgData = document.getElementById('img');
        let tf_img = tf.browser.fromPixels(imgData);
        tf_img = tf_img.expandDims(0);
        const predictions = await this.model.executeAsync(tf_img, ['detection_classes']).then(predictions => {
            const data = predictions.dataSync()
            console.log('Predictions: ', data);
        })

    }

我最终得到了"Error: The output 'detection_classes' is not found in the graph"。我将不胜感激。

Answer 1

this.model.executeAsync(tf_img, ['detection_classes']) 中指定的输出节点可能有错误。另外，这里不需要使用await await this.model.executeAsync(tf_img, ['detection_classes'])。要么使用await，要么使用then。

获取detection_classes 的另一个选项是索引输出数组：

predictions[i].dataSync()[2]