Tensorflow Object Detection: Special NOTE

about matching the IDs in creation of your TFRecord records AND the label_map.pbtxt training file

 

 

 

 

 

• TFRecord Files: For Object Detection API you use TFRecord files to input your data, which basically collects all the images and xml files into a single TFRecord formatted file. This is one of a few ways that data can be presented in Tensorflow and is the primary way for the Object Detection API

• label_map.pbtxt this is a file passed to training in Object Detection API that simply lists out the classes and their ids

  IMPORANT: the ids in label_map.pbtxt should match to the ids used when creating records when creating your TFRecord file. The script I gave you for generating the TFRecord rather than reading in the label_map.pbtxt file has those IDs hard coded. One either fix this and read in from the json formatted label_map.pbtxt file OR make sure the python script matches the label_map.pbtxt.

Example label_map.pbtxt from github covid id mask   item {   id: 1   name: 'face_no_mask' }   item {     id: 2   name: 'face_with_mask' }   item {     id: 3   name: 'face_with_mask_incorrect' }