Skip to content
Other

DiTGS

3D-реконструкция по позам камер

1 min readupdated

DiTGS — исследовательский чекпоинт для 3D-гауссовой реконструкции сцен по изображениям с известными позами камер. Он основан на дообученном трансформере FlashWorld/Wan 2.2 и энкодере TokenGS, а код, установка и обучение вынесены в отдельный репозиторий. Это кастомная PyTorch-модель, а не Diffusers-пайплайн from_pretrained.

Ключевые особенности

  • Восстанавливает 3D-сцену по posed images и intrinsics/extrinsics камер, а не по неразмеченным изображениям.
  • Чекпоинт объединяет дообученный FlashWorld/Wan transformer, замороженные Wan VAE и TokenGS image encoder, дообученный GS decoder/head и новый adapter.
  • Включает независимые DiT-memory attention/KV/LayerScale ветки, кэш пустого промпта и буферы нормализации/rotary.
  • Размер файла — около 22.055 ГБ; для инференса нужен CUDA GPU, но минимальные требования к VRAM в карточке не установлены.
  • Лицензия CC BY-NC-SA 4.0: только некоммерческие академические исследования.

Пример

hf download qyk21/DiTGS --local-dir weights/ditgs
bash scripts/infer.sh 1 --checkpoint weights/ditgs/model.pt --data-root /path/to/DL3DV/benchmark --output outputs/residual/dl3dv_1view
ПодробнееЧекпоинт — raw final main-model после 10 000 warmup и 20 000 joint-обновлений; это не EMA и не ablation-чекпоинт.
Бенчмарк DL3DV: 2-view индекс содержит 140 сцен и 50 целевых кадров на сцену; 1-view берёт первый контекстный кадр.
Инференс по умолчанию: 8 DiT camera slots, Euler NFE=1, seed=42, 256x448, без test-time optimization.
CUDA-проверки скорости и качества при публикации не перезапускались; минимальная VRAM не установлена.
Discuss Telegram 1

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 21 GB · ready files23 GB
  • FP8 / INT8recommendedweights 10 GB12 GB
  • 4-bitweights 5.8 GB6.9 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 5070 · BF16 — from RTX 4090

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.

    # qyk21/DiTGS
    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 qyk21/DiTGS --local-dir ~/models/DiTGS
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to DiTGS
  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 Wan2.2-TI2V-5B-Diffusers (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 1All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 10All on Hugging Face

About the model

qyk21/DiTGS
Author
qyk21
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

01

Ask AI about this model

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

Comments

Video

NAVA

NAVA
Video

MoPlan

MoPlan