The current data loading method poses a risk of memory leakage

[yjb6]

Currently, the data loading method, as shown in the following line, may pose a risk of memory leakage:

LaRa/dataLoader/gobjverse.py

Line 129 in c32ae77

img = np.array(scene[f'image_{view_idx}'])

When I use a similar approach, i.e., obtaining the handle of the HDF5 file during dataset initialization and loading data iteratively, I noticed that my memory usage keeps increasing. I analyzed that this is because the handle is not closed, causing the data to remain in memory. Using the with statement to load data can resolve this issue.

[apchenstu]

Thanks for pointing out this, it would be very helpful if you could submit a pull request, thank you!