Skip to content
Other

WristDepth: metric depth for robot wrist cameras

2 min readupdated

The WristDepth model solves the problem of accurately determining scene depth in meters from a single image and focal length, specifically optimized for cameras on robotic manipulators at distances from 5 cm to 1.5 m. It is based on the Depth Pro 2 B architecture (0.1B parameters in size), fine-tuned on synthetic data for robotics specifics. Unlike the base model, which tends to overestimate distance several times over, WristDepth significantly reduces errors on real-world scenes and outputs an accurate depth map. The model will be useful for robotics engineers and researchers working with computer vision systems.

Key features

  • Robot arm specialization: fine-tuned on synthetic data taking into account gripper visibility in the frame.
  • Error reduction: decreases error on unseen scenes compared to the base version.
  • Ready formats: besides PyTorch weights, provides ONNX graphs and TensorRT support for inference.
  • Metric output: outputs depth directly in meters accounting for the camera's horizontal focal length.

What you can do

  • Robotics integration: apply the model in a robot arm control pipeline to determine distance to objects.
  • Export to ONNX/TensorRT: run optimized versions of the model in FP32, FP16, or INT8 formats for fast inference.
  • Stream processing: use configurations with streaming and clips to analyze frame sequences.

Example

from depth_pro_2 import DepthPro2, load_rgb
import torch

model = DepthPro2.from_pretrained("christianengel/WristDepth", precision=torch.bfloat16).to("cuda").eval()
image, _ = load_rgb("wrist.jpg")
depth = model.infer(image, f_px=450.0)["depth"]
Read moreDetails
Benchmarks: on the DROID test set, the AbsRel error is reduced from 1.445 to 0.255, and the scale is improved from 2.27 to 1.00.
Limitations: scale may vary depending on the setup, calibration for a specific camera is recommended.
Training details: 4,000 frames from BlenderProc procedural renders, self-distillation, and log-depth L1 + gradient matching loss function.
Discuss Telegram 0

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 0.4 GB · ready files0.9 GB
  • BF16weights 0.2 GB0.7 GB
  • FP8 / INT8recommendedweights 0.1 GB0.6 GB
  • 4-bitweights 0.1 GB0.6 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · F32 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # christianengel/WristDepth
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download christianengel/WristDepth --local-dir ~/models/WristDepth
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to WristDepth
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of DepthPro2-B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 2All on Hugging Face
Fine-tunes 4All on Hugging Face

About the model

christianengel/WristDepth
Author
christianengel
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments

Other

Sulphur 3

Sulphur 3