CODEWITHSUNDEEP
pythonobjecttensorflowdetection
Rachna Pathak
Rachna Pathak

Reputation: 155

Training and Validation Accuracy in Tensorflow Object Detection API

I have been using Tensorflow Object detection API on my own dataset. While training, the training losses are updated on the tensorboard. But I need the training and validation accuracy respectively (mAP). What steps need to be taken to get these values?

Upvotes: 1

Views: 1634

Answers (2)

Emanuel Huber
Emanuel Huber

Reputation: 151

Since you said mAP which stands for Mean Average Precision, you need to have the metrics_set in your pipeline config file with the value "coco_detection_metrics". Your file should have something like that:

    eval_config: {
        metrics_set: "coco_detection_metrics"
        use_moving_averages: false
    }

After that, when you run the eval_continuously you should get the mAP on your validation set. For the training set you need to set the eval_on_train_data parameter when running the model_main_tf2.py script.

Upvotes: 0

francoisr
francoisr

Reputation: 4595

If you are using the keras API, through tf.keras, you can add evaluation functions as metrics in the model.fit function. Checkout the official documentation for a list of all available metrics.

You might be interested be interested in tf.metrics.average_precision_at_k. If it doesn't do exactly what you need, you can also implement a custom metric.

Upvotes: 1

Related Questions