Poor results on ZipNeRF for DTU dataset

[HarshaMupparaju]

I tried implementing ZipNeRF for DTU dataset (using the already existing DTU dataloader in this repo). I have found the results to be very poor compared to 360v2 and NeRF-Synthethic. The configs and results are as below. Would appreciate any suggestions to make ZipNeRF work for DTU dataset or any reasons why it wouldn't work for DTU dataset.

Configs: `Config.exp_name = 'test' Config.dataset_loader = 'dtu' Config.near = 0.5 Config.far = 5 Config.factor = 0 Config.hash_decay_mults = 10 Config.dtu_light_cond = 3

Model.raydist_fn = None

PropMLP.disable_density_normals = True PropMLP.disable_rgb = True PropMLP.grid_level_dim = 1

NerfMLP.disable_density_normals = True`

Results of test/novel view for scan8 of DTU with light_condition=3 using the above configs is as below, the train views are being rendered perfectly: