TensorFlow object detection API - Waymo Dataset

General Discussion
,
#1

Hi,

I am currently having issues regarding the use of Waymo datasets in the TensorFlow Object Detection API.

A frame (tfrecord file) contains images from five perspectives (left, front-left, front, front-right and right) with their respective annotations. However, for my project I only need the images and annotation from the front. I have searched the internet but unfortunately have not found any examples to this problem.

I only found the following project. But this one uses all five perspectives. GitHub - gyes00205/waymo_tf_object_detection: Waymo Open Dataset: Tensorflow 2 Object Detection API

I am grateful for any help or ideas.

#3

Hi @muyigl00, To get the data related to camera_Front you can use .map function on the dataset to get data related to the camera_FRONT. For example

dataset_camera_Front = dataset.map(lambda x: x['camera_FRONT'])
for data in dataset_camera_Front:
    image = data['image']
    labels = data['labels']

Thank You.