Strange depth map patterns

I ran the basic evaluation code.
`
```python
import glob, os, torch
from depth_anything_3.api import DepthAnything3
device = torch.device("cuda")
model = DepthAnything3.from_pretrained("depth-anything/DA3NESTED-GIANT-LARGE")
model = model.to(device=device)
example_path = "assets/examples/SOH"
image_paths = sorted(glob.glob(os.path.join(example_path, "*.png")))
prediction = model.inference(
image=image_paths,
#process_res=1208,
export_dir="./output",
export_format="mini_npz-glb"
)
# prediction.processed_images : [N, H, W, 3] uint8 array
print(prediction.processed_images.shape)
# prediction.depth : [N, H, W] float32 array
print(prediction.depth.shape)
# prediction.conf : [N, H, W] float32 array
print(prediction.conf.shape)
# prediction.extrinsics : [N, 3, 4] float32 array # opencv w2c or colmap format
print(prediction.extrinsics.shape)
# prediction.intrinsics : [N, 3, 3] float32 array
print(prediction.intrinsics.shape)
```
`

Everything goes well but the visualization results are strange:


Reproduce the report with the Python snippet, the assets/examples/SOH images, and the DA3NESTED-GIANT-LARGE model, then inspect the files written to ./output in mini_npz-glb format. Compare the exported depth map with prediction.depth and identify whether the strange pattern originates during inference or visualization; the issue is resolved when the cause and expected output are established.

Written by the indexing model from the issue text.