Search code examples
tensorflowobject-detectiontensorboard

Is it necessary to copy images and annotations to Cloud bucket when training Object Detection model?

I am training my own dataset, and have successfully trained a few before. however I have been copying the Images and Annotations folder to the cloud bucket as well, is this necessary? I already have all the TF record and config files, are images/annotations necessary to have in the bucket?

my assumption is the images are necessary, because when running tensorboard to view images and steps, tensorboard needs the image to display?

is this correct? thanks.

this image shows, that the images are inactive. also, there is no PRECISON.MAP chart. data

Solution

  • you don't need to move your images and annotations folder to the cloud bucket. When creating the tfrecords, the images is included. For example, for the Pascal dataset, you can see it here. 

    'image/encoded': dataset_util.bytes_feature(encoded_jpg)