Skip to content
Other

ReTokenGS-stream-dl3dv

Streaming 3D Scene Reconstruction from Video Frames

2 min readupdated

The ReTokenGS-stream-dl3dv model solves the problem of fast 3D scene reconstruction from real-time incoming frames in a feed-forward mode (without additional per-scene optimization during testing). The architecture is based on a closed-loop system with external RMS norm (root mean square normalization) and latent gradient steps, fine-tuned on the DL3DV dataset. The output is a high-quality 3D representation of the scene with support for sequentially arriving views. This model will be useful for researchers and developers in computer vision and neural rendering.

Key features

  • Streaming processing: views can arrive gradually in batches (for example, 8 + 4 + 4).
  • No test-time optimization: fast scene generation without heavy fine-tuning.
  • Trained on the DL3DV dataset with a fixed 50-frame window.

What you can do

  • Reconstruct 3D scenes from sequentially incoming camera angles.
  • Integrate the model into streaming neural rendering and video processing pipelines at a resolution of 256×256.

Example

The model accepts positioned incoming images and processes them through an iterative hidden state update loop. Detailed launch code and comparisons with baseline versions are available in the project repository on the stream_dl3dv branch.

Read moreDetails
• Architecture: loop with latent gradient step and external RMS norm
• Metrics (DL3DV-140, 16 views): PSNR 29.70 / SSIM 0.895 / LPIPS 0.176
• License: Apache-2.0 (based on LocusGS / TokenGS)
• Weight hash: 8aaa60c109ae6e5b7b869e0671278896de141fdc4fdde208d8989d4b3cd8c574
Discuss Telegram 0

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 1.4 GB · ready files2.0 GB
  • BF16weights 0.7 GB1.3 GB
  • FP8 / INT8recommendedweights 0.3 GB0.9 GB
  • 4-bitweights 0.2 GB0.7 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.

    # Yusin/ReTokenGS-stream-dl3dv
    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 Yusin/ReTokenGS-stream-dl3dv --local-dir ~/models/ReTokenGS-stream-dl3dv
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to ReTokenGS-stream-dl3dv
  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 ReTokenGS-stream-dl3dv (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

Yusin/ReTokenGS-stream-dl3dv
Author
Yusin
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