synapse-sr: Sentinel-2 at 2 m, with every pixel accounted for

Weights for synapse-sr, an open-source package that super-resolves Sentinel-2 L2A imagery from 10 m to a 2 m output grid (B04 B03 B02 B08). A physics model of the Sentinel-2 sensor keeps every output consistent with the measurement; every pixel carries a support label (measured vs inferred) and a calibrated uncertainty. Docs: https://sharadhnaidu.github.io/synapse-sr/

Model	Folder	Size	Speed (1.28 km scene)	Use
SYNAPSE Flash v1 (default)	Flash/	12 MB, 0.6 M parameters at inference	~1 s on a laptop CPU	anywhere: CPU, laptops, Colab, Kaggle, ARM
SYNAPSE Pro v2	Pro/	58 MB, 14.4 M parameters	~5 s on a GPU	most detail, GPU

_{RV University, Bengaluru: Sentinel-2 L2A 10 m (left) and synapse-sr 2 m (right).}

Use

pip install synapse-sr

import synapse_sr

r = synapse_sr.super_resolve("sentinel2_l2a.tif")                  # Flash; model="pro" for Pro
r.save("sentinel2_2m.tif")                                         # georeferenced GeoTIFF
r.uncertainty()                                                    # calibrated expected error per pixel

synapse-sr --fetch 12.92,77.50 --dates 2025-01-01:2025-03-15 out.tif --preview preview.png

Weights download once and are SHA-256 verified. Offline: Flash.from_pretrained(weights="Flash/synapse-flash-v1.safetensors").

Benchmark

Official opensr-test protocol, mean over NAIP, SPOT, Spain urban, Spain crops and VENuS (178 scenes); package defaults.

Model	Improvement ↑	Omission ↓	Hallucination ↓	Detail corr. ↑	RMSE ↓	Spectral error ↓	Reflectance error ↓
SYNAPSE Flash	0.148	0.756	0.096	0.298	0.0237	0.427	0.0019
SYNAPSE Pro	0.152	0.750	0.098	0.297	0.0237	0.428	0.0019
SEN2SR	0.150	0.759	0.091	0.284	0.0235	0.665	0.0025
SEN2SR-Lite	0.152	0.749	0.099	0.290	0.0234	0.463	0.0019
LDSR-S2	0.197	0.599	0.204	0.206	0.0240	1.015	0.0036
Satlas ESRGAN	0.129	0.181	0.690	0.089	0.0443	7.787	0.0242
Bicubic	0.102	0.830	0.068	0.279	0.0234	0.601	0.0028

How it works

x_hat = x_base + P_N(delta): x_base is a regularised inversion of the exact Sentinel-2 sensor model (per-band point-spread function); delta is predicted by the network; P_N projects it onto the part of the image the sensor cannot see, so the network cannot change what the satellite measured.

• Pro: Mamba state-space backbone (6 x 8 visual state-space blocks), 20 m spectral-context stem, frequency mixer, direct x5 PixelShuffle head.
• Flash: re-parameterised convolutional network on all ten bands (trains as three-branch convolutions, folds to one 3x3 convolution per layer), trained by knowledge distillation from Pro.

Training data

• Sentinel-2 L2A paired with 0.6 m NAIP aerial imagery (951 real training pairs, United States).
• A streamed corpus of 3,044 native 1 m NAIP patches across 15 land-cover classes, observed through the exact Sentinel-2 sensor model, with composed structures (roads, buildings, field and water edges).
• ISRO Cartosat-2E and Cartosat-3 imagery (sample data from NRSC's Bhoonidhi portal): 1,001 Indian training tiles from three scenes, plus a separate held-out Cartosat-3 scene, observed through the Sentinel-2 sensor model.

Calibrated uncertainty

Measured coverage on held-out development pixels at the 80 / 90 / 95 % levels: Flash 82 / 91 / 95 %, Pro 82 / 91 / 96 %.

Files

File	Content
Flash/synapse-flash-v1.safetensors	Flash weights (model.*) and the Sentinel-2 operator kernels (operator.weight)
Pro/synapse-pro-v2.safetensors	Pro v2 weights and operator kernels
Pro/synapse-pro-v1.safetensors	Pro v1, kept for reproducibility
assets/	before / after examples produced with the package

Examples

RV University, Bengaluru	Bengaluru city centre
Ludhiana, Punjab: fields	Wayanad, Kerala: landslide-affected hills

Licence and acknowledgements

CC0-1.0. Third-party components and their licences are listed in THIRD_PARTY_NOTICES in the package. Sentinel-2 data: Copernicus programme, European Space Agency. Cartosat data: ISRO / NRSC (Bhoonidhi).