xFormers not available → model imports OK, but infer_image returns all-zero depth map (min=0.0, max=0.0)

[Wasiq1123]

Hi @heyoeyo — thanks for the help earlier. Uninstalling xFormers removed the import error, but now I'm seeing the model produce an all-zero depth map.

### Environment

• OS: Ubuntu 22.04
• Python: 3.10
• Device: CPU only (no CUDA; torch.device -> 'cpu')
• xFormers: uninstalled (or disabled via XFORMERS_DISABLE_MEMORY_EFFICIENT_ATTENTION=1)
• Repo version: Depth-Anything-V2 (local copy)

### Encoder / checkpoint

• Encoder used: vits (I also tested vitb, vitg variants)
• Checkpoint: metric_depth/checkpoints/depth_anything_v2_metric_hypersim_vits.pth (loaded with map_location='cpu')

### Minimal repro (my code)

import cv2
import torch
from depth_anything_v2.dpt import DepthAnythingV2

DEVICE = 'cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu'

model_configs = {
    'vits': {'encoder': 'vits', 'features': 64, 'out_channels': [48, 96, 192, 384]},
    'vitb': {'encoder': 'vitb', 'features': 128, 'out_channels': [96, 192, 384, 768]},
    'vitl': {'encoder': 'vitl', 'features': 256, 'out_channels': [256, 512, 1024, 1024]},
    'vitg': {'encoder': 'vitg', 'features': 384, 'out_channels': [1536, 1536, 1536, 1536]}
}

encoder = 'vits'
dataset = 'hypersim'
model = DepthAnythingV2(**model_configs[encoder])
checkpoint_path = f'/path/to/checkpoints/depth_anything_v2_metric_{dataset}_{encoder}.pth'
model.load_state_dict(torch.load(checkpoint_path, map_location='cpu'))
model = model.to(DEVICE).eval()

image_path = "/home/wasiq/Pictures/2m_Depth_Distance.jpeg"
raw_img = cv2.imread(image_path)
depth = model.infer_image(raw_img)
print(f"The minimum depth is {depth.min()}")
print(f"The maximum depth is {depth.max()}")

**### Observed behavior**
xFormers not available printed (twice in logs)
The minimum depth is 0.0
The maximum depth is 0.0

**### Expected behavior**
Non-constant depth map with varied values (for my test image the ground-truth near a marked point is ~2.0 m)
depth.min() < depth.max() and meaningful spatial variation

**### Diagnostics I already tried**
Confirmed checkpoint loads without crash (no obvious exceptions) but I haven't validated state_dict key names yet.
Tried different encoder names (vits, vitb) consistent with checkpoint filenames.
Ensured XFORMERS_DISABLE_MEMORY_EFFICIENT_ATTENTION=1 or uninstalled xFormers.

[gururajbg]

I tested this on Windows 11 with an RTX 3050 (6GB), Python 3.8 (conda), and PyTorch 2.4.1 + CUDA 11.8.

GPU inference works correctly even when xFormers is not available.
Using both run.py and a minimal infer_image() script, the depth output is non-constant
(e.g. min=0.0, max≈816).

When explicitly forcing CPU inference (device="cpu"), inference fails with:

RuntimeError: Input type (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same

This suggests that during CPU inference, the input tensor is still created on CUDA
inside the preprocessing / infer_image() path, while the model weights are on CPU.

This may explain CPU-only failures or zero-depth outputs reported in this issue.

[gururajbg]

Follow-up: I tested a minimal fix locally.

In infer_image() (depth_anything_v2/dpt.py), moving the input tensor to the model’s device fixes CPU inference:

image = image.to(next(self.parameters()).device)

With this change, CPU-only inference runs successfully and produces non-constant depth
(e.g. min=0.0, max≈816 on my setup).

This confirms the issue is due to a device mismatch during preprocessing rather than xFormers itself.

[sourcesync]

Hey @gururajbg @Wasiq1123 I am also seeing all zeros. I've tried setting device to cpu explicitly as you did above, but I'm still seeing all zeroes. Any thoughts how I can move past this?

[gururajbg]

@sourcesync I saw similar behaviour while testing this.

If the device fix didn’t help, check you’re using the metric model import with the metric checkpoint — mixing that up can give zeros. Also just make sure the image actually loaded (cv2 sometimes returns None silently).

In my case it was device mismatch, but zeros can also come from model/checkpoint mismatch.