Skip to content
Images

Photo Denoising Autoencoder

простой автоэнкодер для очистки фото от шума

2 min readupdated

Лёгкая свёрточная модель (около 2 млн параметров), которая убирает синтетический гауссовский шум с цветных фотографий. Энкодер сжимает изображение вчетверо по ширине и высоте, а декодер восстанавливает чистый снимок из этого сжатого представления без skip-connections (прямых пробросов признаков между слоями). Модель создана автором nisarganag в рамках курсовой работы по глубокому обучению и подойдёт студентам или разработчикам, которым нужен наглядный и компактный пример работы denoising-автоэнкодера в PyTorch.

Ключевые особенности

  • Компактная архитектура на 1 987 203 параметра, обученная на 400 фотографиях из датасета BSDS500.
  • Работает с аддитивным белым гауссовским шумом уровня σ 5–50 по шкале 0–255.
  • На тесте CBSD68 заметно обходит классические фильтры вроде Gaussian blur, Median filter и Non-local means.
  • Веса и код модели выложены под свободной лицензией MIT, также доступно интерактивное демо.

Что можно сделать

  • Использовать как готовый наглядный пример для изучения работы свёрточных автоэнкодеров и задач image-to-image на PyTorch.
  • Встроить в учебные или исследовательские пайплайны для очистки цветных снимков от синтетического гауссовского шума.
  • Взять за базовый бейзлайн при экспериментах с добавлением skip-connections или обучением на других типах искажений.

Пример

from huggingface_hub import hf_hub_download
hf_hub_download("nisarganag/photo-denoising-autoencoder", "model.py", local_dir=".")
import numpy as np
from PIL import Image
from model import DenoisingAutoencoder, add_gaussian_noise, denoise, image_to_tensor, tensor_to_image

model = DenoisingAutoencoder.from_pretrained("nisarganag/photo-denoising-autoencoder").eval()
clean = image_to_tensor(np.array(Image.open("photo.png").convert("RGB")))
noisy = add_gaussian_noise(clean, sigma=25)
Image.fromarray(tensor_to_image(denoise(model, noisy))).save("cleaned.png")
ПодробнееМетрики на CBSD68 (PSNR / SSIM): 32.97 dB / 0.922 при σ=15, 30.65 dB / 0.876 при σ=25 и 27.41 dB / 0.776 при σ=50.
Обучение: 17 эпох (46,9 минуты) на Apple M4 GPU (PyTorch MPS) с функцией потерь MSE и оптимизатором Adam.
Ограничения: из-за сжатия вчетверо слегка сглаживает мелкие текстуры и тонкие линии; не предназначена для реального шума камер, JPEG-артефактов, шума «соль и перец», медицинских или криминалистических снимков.
Discuss Telegram 0 40

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 0.0 GB · ready files1.5 GB
  • BF16weights 0.0 GB1.5 GB
  • FP8 / INT8recommendedweights 0.0 GB1.5 GB
  • 4-bitweights 0.0 GB1.5 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.

    # nisarganag/photo-denoising-autoencoder
    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 nisarganag/photo-denoising-autoencoder --local-dir ~/models/photo-denoising-autoencoder
    # The model is in ~/models. How to run it is in the model card (link above).
  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.

  • SageAttention 3v2.2.0from October 28, 2025

    even faster on RTX 50 / Blackwell

    FP4 attention for Blackwell. Faster than SageAttention 2, but the authors suggest v2 when accuracy matters.

    How to enable: Build from the sageattention3_blackwell folder of the repo; in ComfyUI via KJNodes.

    Earlier generations (2)
    • SageAttention 2++ — Quantised (INT8/FP8) attention with no visible quality loss. The main speed-up for image and video generation.
    • SageAttention 1 — First version: INT8 for Q·K. Still works, but 2++ is faster and more accurate.
  • 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 (3)
    • 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.
    • xFormers — The older way to speed up attention, works almost everywhere. Sage or Flash are faster where available.
  • cache-dit (DBCache)v1.5.6from November 20, 2026new release

    1.5–3× at nearly the same quality

    Reuses block outputs between diffusion steps. Supports most DiT models.

    How to enable: pip install cache-dit; one line cache_dit.enable_cache(pipe) for diffusers.

    Earlier generations (2)
    • MagCache — Skips steps by magnitude of change. More accurate than TeaCache, no calibration.
    • TeaCache — The first popular diffusion step cache. Works, newer methods are more accurate.
  • Nunchaku (SVDQuant 4-bit)v1.2.1from January 25, 2026

    ≈3× faster and 3.5× less memory

    4-bit versions of popular image models with almost no quality loss (FP4 on RTX 50).

    How to enable: Ready weights for supported models; in ComfyUI use ComfyUI-nunchaku.

  • 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 photo-denoising-autoencoder (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

nisarganag/photo-denoising-autoencoder
Author
nisarganag
Category
Images
Released
Posted

On NeyroLearn

Hugging Face

400

Ask AI about this model

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

Comments

Other

AI Photo

AI Photo