I think I may have solved the issue, the dataset_convert specs file was missing the class mapping of some of the classes in my dataset which caused the tfrecord files to be generated incorrectly.
kitti_config {
root_directory_path: “/workspace/tao-experiments/data/”
image_dir_name: “image”
label_dir_name: “label”
image_extension: “.png”
partition_mode: “random”
num_partitions: 2
val_split: 0
num_shards: 10
}
image_directory_path: “/workspace/tao-experiments/data/”
target_class_mapping {
key: “object”
value: “object”
}
target_class_mapping {
key: “remote”
value: “remote”
}
target_class_mapping {
key: “cell_phone”
value: “cell_phone”
}
target_class_mapping {
key: “toothbrush”
value: “toothbrush”
}