Skip to content
Other

PaleVision: An evolutionary approach to spectral vision

2 min read

A new research model from NeurIPS 2026 explores the evolution of color vision (EVPL — evolutionary visual perception layers) for semantic segmentation tasks on hyperspectral images. The solution models the spectral responses of opsins and combines them with convolutional networks and segmentation heads to handle complex spectral data. The development targets computer vision and spectral data processing researchers who need to reproduce experiments or use pre-trained weights to analyze specific datasets.

Key features

  • EVPL architecture: The use of trainable spectral response layers based on biological models of vision evolution.
  • High metrics: The availability of pre-trained state-of-the-art weights for the MODID, MHSI, and MinnApple datasets with fixed mean intersection over union results.
  • Configuration flexibility: Support for various backends, including MiT-B0, Twins-SVT-Small, and PVT-v2-B2.
  • Training tools: PyTorch and torchrun scripts for multi-process experiment execution.

What you can do

  • Mars analysis: Run training and evaluation of spectral filters and segmentation on Mars MSL data.
  • Agricultural data processing: Apply checkpoints for detection, counting, and segmentation on the MinnApple dataset.
  • Encoder testing: Perform encoder swaps (encoder_swap_exp.py) with customizable opsin modes.
  • Weight inference: Load pre-trained checkpoints (backbone, head, and EVPL) in Python scripts for testing on custom data.

Example

import torch
ckpt = torch.load("weights/modid_twins_svt_small_free.pth", map_location="cpu")
backbone.load_state_dict(ckpt["backbone"])
head.load_state_dict(ckpt["head"])
evpl.load_state_dict(ckpt["evpl"])
Read moreDetails
Benchmarks: MODID (mean intersection over union 0.6006), MHSI (mean intersection over union 0.5727), MinnApple segmentation (mean intersection over union 0.4845), MinnApple counting (mean absolute error 2.446).
Hardware: An NVIDIA graphics processing unit with sufficient memory for backends at config resolutions; Central Processing Unit loading is supported.
Limitations: Weights contain only module parameters without optimizer states; models are tightly coupled to their training datasets.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.2 GB · ready files0.7 GB
  • FP8 / INT8recommendedweights 0.1 GB0.6 GB
  • 4-bitweights 0.0 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 · 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.

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

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

About the model

smilelab/PaleVision
Author
smilelab
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

HMH-1-Cell

HMH-1-Cell