Support datasets with images with varied widths and heights

Hi the project is very interesting.

Are you able to add support for datasets with images with varied widths and heights?

Error executing job with overrides: ['train.gs_epochs=30000', 'train.no_densify=True', 'gs.dataset.source_path=data/frame000001', 'gs.dataset.model_path=data/output/frame000001', 'init_wC.matches_per_ref=20000', 'init_wC.nns_per_ref=3', 'init_wC.num_refs=180', 'wandb.mode=disabled']
Traceback (most recent call last):
  File "/mnt/c/Users/ice/wslworkspace/EDGS/train.py", line 54, in main
    trainer.init_with_corr(cfg.init_wC)
  File "/mnt/c/Users/ice/wslworkspace/EDGS/source/trainer.py", line 234, in init_with_corr
    camera_set, selected_indices, visualization_dict = init_fn(
  File "/mnt/c/Users/ice/wslworkspace/EDGS/source/corr_init.py", line 630, in init_gaussians_with_corr
    kptsA_np, kptsB_np, kptsB_proj_matrices_idcs, kptsA_color, kptsB_color = extract_keypoints_and_colors(
  File "/mnt/c/Users/ice/wslworkspace/EDGS/source/corr_init.py", line 362, in extract_keypoints_and_colors
    imB_compound_np = np.stack(imB_compound, axis=0)
  File "/home/ice/miniconda3/envs/edgs/lib/python3.10/site-packages/numpy/core/shape_base.py", line 449, in stack
    raise ValueError('all input arrays must have the same shape')
ValueError: all input arrays must have the same shape

Set the environment variable HYDRA_FULL_ERROR=1 for a complete stack trace.

Start in source/corr_init.py at extract_keypoints_and_colors, especially the np.stack call around line 362, and trace its inputs from init_gaussians_with_corr. Reproduce the failure using the training command in the report. Done means the training initialization accepts images with varied widths and heights without the reported shape error.