Skip to content
Computer vision

qwen3d_points_dist - Qwen-3D fine-tune for indoor change detection using point clouds

2 min readupdated

An experimental model based on Qwen-3D (3B) is designed to detect changes between an initial room point cloud and a new RGB-D frame (an image with depth data). It identifies which objects were moved or deleted, returning the output in JSON format along with the movement distance in meters. The architecture voxelizes the original points and uses a specialized voxel encoder combined with the visual transformer of the base model via 4D RoPE. Trained using LoRA (a lightweight adapter added to the base model), this model is useful for spatial intelligence and robotics researchers studying indoor environment changes.

Key features

  • Based on Qwen-3D (3B) and trained on the vlm-change-detection-4k dataset.
  • Uses a voxel encoder to encode points and color from the original point cloud.
  • Applies LoRA with rank r=16 to adapt the weights of both the LLM and the encoder.
  • Outputs a structured JSON response detailing actions (moved, deleted) and distance metrics.

What you can do

  • Apply to spatial analysis tasks and indoor change monitoring using sensor data.
  • Compare an initial room map against a new RGB-D frame within research pipelines.
  • Integrate with the code from the pointcloud-voxel-fusion branch of the spatial_llm repository.
Read moreDetails
Initialization: katefgroup/Qwen-3D qwen3d_3b.pth
Training: LoRA r=16, alpha=32, 15,000 iterations on 1x NVIDIA RTX 5090 (~6.5 h)
Limitations: slow initial startup, encoder trained from scratch, large rooms lose some voxels due to the 6,000 token limit
License: CC BY-NC 4.0 (non-commercial use)
Discuss Telegram 0

Will it run?

Runs in FP8 / INT8

Full precision does not fit, a quantised version does: quality barely changes down to 8 bits, more noticeably at 4 bits.

  • BF16weights 24 GB · ready files26 GB
  • FP8 / INT8recommendedweights 15 GB18 GB
  • 4-bitweights 12 GB14 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 4090 · BF16 — from RTX 5090

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

This is a LoRA: counted together with its base model Qwen-3D.

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.

    # jimtjames/qwen3d_points_dist
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    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 katefgroup/Qwen-3D --local-dir ~/models/Qwen-3D
    hf download jimtjames/qwen3d_points_dist --local-dir ~/models/qwen3d_points_dist
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Qwen-3D
  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

Requires the base model

katefgroup/Qwen-3D

This model is a LoRA / adapter. Download the base model first, then load this one on top.

New posts about versions of Qwen-3D (quants, finetunes, LoRAs) will show up under the bell.

LoRAs and adapters 2All on Hugging Face

About the model

jimtjames/qwen3d_points_dist
Author
jimtjames
Category
Computer vision
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