nikhilr01/sen-bridge-npy

SEN-Bridge, memory-mappable .npy release

The same 170,000 radar-to-optical pairs as nikhilr01/sen-bridge (125,000 train / 15,000 validation / 30,000 test; see that card for how the data were built, filtered and split), stored as raw 8-bit arrays so that random access is a slice of a memory-mapped file with no image decoding. It is an optional extra copy; the Parquet release remains the reference. The pixels are identical to the decoded PNGs of that release (checked on every row by comparing arrays, see below).

Files

File	Contents
<split>_sar.npy	uint8 [N, 128, 128, 3], R=VV, G=VH, B=VV-VH
<split>_optical.npy	uint8 [N, 128, 128, 3], RGB (B4, B3, B2)
<split>_meta.npy	float32 [N, 7] conditioning vectors
<split>_ids.npy	[N] example ids
metadata.parquet	the other columns of the main release, plus row
MANIFEST.sha256	SHA-256 of every file

<split> is train, validation or test. Row i of the arrays of a split is row i of its Parquet shards (files in name order) and the row with row == i and that split in metadata.parquet, whose split column uses these names (the main release writes val there for validation). meta is (sin doy, cos doy, x, y, z, dt / 48 h, orbit), the orbit being +1 (descending) or -1 (ascending).

Loading

With the sen-bridge package, which downloads from a pinned revision and checks every file:

from sen_bridge import data

train = data.load_dataset("train")          # or "validation", "test"
pair = train[1234]                          # sar, optical, meta, id
batch = train[[5, 1, 9]]                    # arrays with a leading batch axis
for batch in train.iter_batches(256, shuffle=True, seed=0):
    ...

A Split has __len__ and __getitem__, so it works as a map-style dataset in PyTorch and Grain, and it pickles by path so worker processes reopen the memory maps. load_dataset("train", float_images=True) returns float32 images in [-1, 1]. Without the package:

import numpy as np

sar = np.load("train_sar.npy", mmap_mode="r")   # read after `hf download`

Size: about 16.7 GB of arrays. Reads are fastest from a local SSD with the files in the page cache.

License and attribution

CC BY 4.0 (LICENSE), the licence of the source dataset SSL4EO-S12 v1.1. Contains modified Copernicus Sentinel data 2019-2021. Attribution, changes made and the third-party notices are in NOTICE and in the main dataset card.

Citation

@misc{raghavendra2026senbridge,
  title  = {{SEN-Bridge}: A SAR-to-Optical Flow Matching Dataset},
  author = {Raghavendra, Nikhil},
  year   = {2026},
  doi    = {10.57967/hf/11204},
  url    = {https://huggingface.co/datasets/nikhilr01/sen-bridge}
}
@article{blumenstiel2025ssl4eos12v11,
  title   = {{SSL4EO-S12 v1.1}: A Multimodal, Multiseasonal Dataset for
             Pretraining, Updated},
  author  = {Blumenstiel, Benedikt and Ait Ali Braham, Nassim and
             Albrecht, Conrad M and Maurogiovanni, Stefano and Fraccaro, Paolo},
  journal = {arXiv preprint arXiv:2503.00168},
  year    = {2025}
}
@article{wang2022ssl4eos12,
  title   = {{SSL4EO-S12}: A Large-Scale Multi-Modal, Multi-Temporal Dataset for
             Self-Supervised Learning in Earth Observation},
  author  = {Wang, Yi and Ait Ali Braham, Nassim and Xiong, Zhitong and
             Liu, Chenying and Albrecht, Conrad M and Zhu, Xiao Xiang},
  journal = {arXiv preprint arXiv:2211.07044},
  year    = {2022}
}