doem1997/cs701-sar-course-data

CS701 SAR course data

SAR images for the CS701 team assignment (SMU, AY2026-27 Term 1): predict the class of each image (one of 9) and detect its objects. Rules, timeline and submissions: Codabench. Reference code: GitHub.

Releases

split	images	labels and boxes	files	released (SGT)
train	9,392	included (32,533 boxes)	train.tar.gz	30 Sep 2026
val	1,426	withheld	val.tar.gz, sample_submission_val.zip	12 Oct 2026, 00:00 (Phase 1 opens)
test	2,065	withheld	test.tar.gz, sample_submission_test.zip	2 Nov 2026, 00:01 (Phase 2 opens)

Val and test labels are never released: hold out part of train to validate.

Download

pip install -U huggingface_hub
hf download doem1997/cs701-sar-course-data --repo-type dataset --local-dir cs701-sar-course-data
cd cs701-sar-course-data && tar -xzf train.tar.gz

After each release, run hf download again and unpack the new archive.

Format

• classes.json: aircraft, airport, bridge, car, harbor, oil_tank, playground, ship, wind_turbine. label 0-8 is the class of the image; category_id = label + 1 is the class of a box.
• train/labels.csv: image_id,file_name,label, one row per image.
• train/instances.json: the boxes in COCO format, bbox = [x, y, w, h] in pixels of the original image.
• val/images.json, test/images.json: the images to predict (ids, file names, sizes).
• Every image has one class label and at least one box, all of that class. The images are 8-bit gray, mostly 512 × 512 or 256 × 256, JPEG or PNG; many store the gray values in three channels, so read them with Image.open(path).convert('L') or as RGB.

The submission format is on the Codabench Evaluation page.

Licence and citation

The images and boxes come from SARDet-100K, RSAR, M4-SAR and HRSID, reorganised by SARFact, and keep their original terms. This course release (selection, splits, class labels, files) is licensed CC BY-NC 4.0: use it for non-commercial research and teaching only. Details and the citations of the four datasets, which you should cite: SOURCES_AND_LICENSES.md.

Contact: TA Zichen, zichen.tian.2023@phdcs.smu.edu.sg