0

I have been studying transfer learning with models like inception_v4 and inception_resnet_v2. Found some projects that uses bottleneck and some uses tfrecords to store the training images. When retraining the inception_v4 model with the same data using those two methods bottleneck gave 95% accuracy and tfrecord only gave 75%. But, all the new projects seems to use tfrecords for data and .ckpt format to store the model. Can someone explain me whats the difference and which one is better in which case

Eshaka
  • 974
  • 1
  • 14
  • 38

1 Answers1

1

If you are working with large datasets, using a binary file format for storage of your data can have a significant impact on the performance of your import pipeline. Hence, it will affect your training time of the model.

By using TFRecords, it is possible to store sequence data. For e.g, a series of data. Besides, it easy to combine multiple datasets and integrates seamlessly with the data import and preprocessing functionality provided by the library.

For more information about TFrecords, please refer this link.

Lee
  • 181
  • 1
  • 3
  • 22
  • Thanks for the answer. For now i have a very little data set, around 300 images in total. So which one will work better. does this file format has an effect on the final result(accuracy) or just the speed. – Eshaka Jan 29 '19 at 06:30
  • 1
    For consistency, it is better to use tfrecords. For my experience in object detection model training, the format is in tfrecords and .ckpt , etc. So, it does not affect the accuracy of your model. If you want a better accuracy, you need more dataset. 300 images is too small. – Lee Jan 29 '19 at 06:37
  • i thought the same way.. but for now this is all i have. So need to try my best to get something out of this data. i used this code to train the model to inception v4 with some modifications [link](https://github.com/MicrocontrollersAndMore/TensorFlow_Tut_2_Classification_Walk-through). this one gave 95% accuracy and it used bottleneck. but i ran into some other issues while saving and loading the model after training. So that is how i ended up looking at these new projects with tfrecords. but they give a really bad accuracy. – Eshaka Jan 29 '19 at 06:51