I have trained a model for object detection using tensorflow object detection repository.
Now while evaluating the default way is to read from a record
file using tf.contrib.slim.parallel_reader
I want to read the images directly using file system and get the prediction for the image. How do I do that?
have a look on this https://github.com/tensorflow/models/blob/master/object_detection/object_detection_tutorial.ipynb, here a frozen model is used, but you can use normal ckpt with minor changed logic to use image instead of tfrecord.
step1: Load the model using ckpt file
step2: get the model end points to run predictions; similar to the link above except removing the :0
step3: use placeholder to feed the numpy array (your images)
Apart from this you can convert the trained model into a frozen model convert to frozen modeland follow the same step as described in the above link.