How to read waymo Block-NeRF dataset?

[DRosemei]

I find that waymo Block-NeRF dataset has been provided here. I want to know how to parse these tfrecords.

[VincentCa]

Hi, thank you for your interest in the Block-NeRF dataset! We will add some more instructions to the website shortly. In the meantime, you can follow the example below:

files = tf.io.gfile.glob('waymo_block_nerf_mission_bay_validation.tfrecord*')
dataset = tf.data.TFRecordDataset(filenames=files, compression_type='GZIP')

iterator = dataset.as_numpy_iterator()

for entry in iterator:
  example = tf.train.Example()
  example.MergeFromString(entry)
  
  # Decode RGB image.
  img_data = example.features.feature['image'].bytes_list.value[0]
  rgb = tf.image.decode_png(img_data).numpy()
  print(rgb.shape)

  # Find other entries as described on the website.

[VincentCa]

Hi, you should find that the training split contains about 11,269 and the validation split contains about 576 images. Could you let me know how you counted the number of examples?

[vinodrajendran001]

Hi @VincentCa ,

May I know how to get 360 degree images for any 1 frame from *.tfrecord? To be more precise, for a given frame I want the 8 camera (mounted on the top) images and 4 camera (front) images.

Thanks.

[YZsZY]

Hi @VincentCa ,

May I know how to get 360 degree images for any 1 frame from *.tfrecord? To be more precise, for a given frame I want the 8 camera (mounted on the top) images and 4 camera (front) images.

Thanks.

This dataset provides the camera serial number and camera location for each image, you just need to filter the cameras by serial number and extract the corresponding images