Skip to content
Other

BFL

taxonomy-aware 3D bird shape reconstruction

2 min readupdated

This model tackles 3D body reconstruction of birds from images, recovering proportions and poses more accurately. It is a fine-tuned version of the AniMer+ architecture trained by the authors of a NeurIPS 2026 paper. Its defining feature is an Ornstein-Uhlenbeck taxonomic prior (a mathematical model of evolutionary similarity across species) applied during training. At inference (when generating outputs), it runs just like standard AniMer+ with zero extra overhead. It is useful for computer vision and bioinformatics researchers, as well as 3D creators building wildlife simulations.

Key features

  • Trained on the CtrlAVES3D dataset for 30,000 steps using inter-species taxonomic relationships.
  • The taxonomic prior is applied strictly during training — at inference, it runs as a standard AniMer+ architecture.
  • Includes the main bfl.pt checkpoint alongside ablation study weights (taxonomy_real.pt and taxonomy_permuted.pt).
  • All files are saved via torch.save and contain a dictionary with model_state weights, the training step, and metrics.

What you can do

  • Reconstruct 3D bird meshes across various species on specialized datasets like CUB or CowBird.
  • Reproduce the NeurIPS 2026 paper experiments and benchmarks using the official eval.py evaluation script.
  • Use the checkpoints as a ready-made baseline for downstream 3D animal pose and morphology analysis in computer vision.

Example

python scripts/download_checkpoints.py
python eval.py --ckpt checkpoints/bfl.pt --dataset COWBIRD --out-json outputs/bfl/eval_COWBIRD.json
Read moreDetails
Main bfl.pt checkpoint metrics: CUB AUC — 89.31, CUB PCK@0.1 — 57.91.
On the CowBird benchmark: AUC — 60.91, PCK@0.1 — 19.26; on CtrlAVES3D: AUC — 93.04.
Evaluated using the eval.py script on an RTX 4090 GPU.
Full setup, weight download, and inference code is available on GitHub (rocknroll17/BFL).
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 2.9 GB · ready files3.7 GB
  • FP8 / INT8recommendedweights 1.5 GB2.1 GB
  • 4-bitweights 0.8 GB1.4 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 · BF16 — 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.

    # rocknroll17/BFL
    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 rocknroll17/BFL --local-dir ~/models/BFL
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to BFL
  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 BFL (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

rocknroll17/BFL
Author
rocknroll17
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