how can I get pose.mat?

[chdzhen]

'''
def create_dataloader_callback(sidelength, batch_size, query_sparsity):
train_dataset = RealEstate10k(img_root="data_download/realestate/train",
pose_root="poses/realestate/train.mat",
num_ctxt_views=opt.views, num_query_views=1, query_sparsity=192,
lpips=opt.lpips, augment=(not opt.no_data_aug))

    .....
    val_dataset = RealEstate10k(img_root="data_download/realestate/test",
                                  pose_root="poses/realestate/test.mat",
                                  num_ctxt_views=opt.views, num_query_views=1, augment=False)

'''
pose_root="poses/realestate/train.mat",
pose_root="poses/realestate/test.mat",

[yilundu]

You can find it here: https://www.dropbox.com/s/qo8b7odsms722kq/cvpr2023_wide_baseline_data.tar.gz?dl=0

[chdzhen]

thanks!
can you share the generate *.mat script?

[yilundu]

Yes it should be below:

import glob
from tqdm import tqdm
import numpy as np
from scipy.io import savemat


if __name__ == "__main__":
    files = glob.glob("train/*.txt")

    full_dict = {}

    for file in tqdm(files):
        # file_dict = {}
        data = np.loadtxt(file, skiprows=1)

        # if len(data.shape) == 1:
        #     continue

        # name, entry = data[:, 0], data[:, 1:]

        # for n, e in zip(name, entry):
        #     file_dict[n] = e

        base = file.split("/")[-1].split(".")[0]
        full_dict[base] = data

    savemat("train.mat", full_dict, do_compression=True)
    import pdb
    pdb.set_trace()
    print(files)

[chdzhen]

why HW set to 256, 455?

in realestate10k_dataio.py

class RealEstate10K():
    def init():
    ***
      data = np.load(dummy_img_path)
      key = list(data.keys())[0]
      im = data[key]
      print(im.shape)
      H, W = im.shape[:2]
      H, W = 256, 455
      self.H, self.W = H, W
      self.augment = augment

      self.square_crop = square_crop

      xscale = W / min(H, W)
      yscale = H / min(H, W)

      dim = min(H, W)

      self.xscale = xscale
      self.yscale = yscale

      # For now the images are already square cropped
      self.H = 256
      self.W = 455
    ***