dejanb/colombia

UAV Thermal Imagery of Buried Landmines

Mirror of "Test Images of Buried landmines" (Universidad del Valle), repackaged from .rar to .zip for use in Google Colab.

This is not the original source. If you use this data, cite the original authors and DOI (see Citation). Content is unmodified — only the archive container changed. Why a mirror? Mendeley's download endpoint redirects to an S3 bucket that returns HTTP 403 for Google Cloud IP ranges, so the original link fails inside Colab while working from a personal machine. .zip also avoids needing unrar in the runtime.

Contents

Single archive: DatainBrief_dataset_1.zip (~310 MB extracted, 1834 images).

Thermal and visible-spectrum imagery of landmines buried at different depths, captured with a DJI Matrice 100 + Zenmuse XT thermal camera over a 10×10 m test field divided into 9 zones, across 5 flight sessions in February–March 2020.

Directory layout

DatainBrief_dataset_1/<session>/<modality>/<zone folder>/<image>.jpg

• —<session>: 13_02_2020, 17_02_2020, 20_02_2020, 24_02_2020, 03_03_2020
• —<modality>: TIFF, R_JPG, JPG
• —<zone folder>: e.g. Zone 7 Mine 5cm depth, Zone 6 free zone
• —filenames encode flight altitude and centre-of-frame temperature, e.g. FlightAltitude__3.2m_TemperatureInCenter__36°C.jpg

Modalities

Labels

Labels are implicit in the folder names — there are no annotation files. Each zone folder gives an image-level (weak) label:

There are no bounding boxes or segmentation masks — localisation requires your own annotation.

Known quirks

Worth handling explicitly in any loader:

• —Folder names are inconsistent: Zone 5 Sruface Mine 0cm depth (typo), Zone 6 free zone (double space), 1 cm vs 1cm.
• —15 files have a corrupt flight altitude in the name (FlightAltitude__-0.m, -1.).
• —13_02_2020/JPG/ contains leftover test folders min1_1m … min8_1m, sana_1m with 0–6 images each; exclude them.
• —The class balance is heavily skewed: ~89 % of images contain a mine (the free zone is 1 of 9), so F1 and accuracy are misleading — prefer ROC-AUC, balanced accuracy and specificity.
• —Flight altitude spans 0–11 m, which drives the apparent size of the mine in frame; treat it as a covariate alongside burial depth.

Loading

from huggingface_hub import hf_hub_download
import zipfile

z = hf_hub_download("dejanb/uav-thermal-buried-landmines",
                    "DatainBrief_dataset_1.zip", repo_type="dataset")
with zipfile.ZipFile(z) as f:
    f.extractall("data")

Reading a raw thermal frame (note: no division by 255):

import numpy as np
from PIL import Image

raw = np.array(Image.open(path)).astype(np.float32)   # 16-bit digital counts
lo, hi = np.percentile(raw, 1), np.percentile(raw, 99)
img = np.clip((raw - lo) / (hi - lo), 0, 1)

Citation

Cite the original dataset:

@misc{tenorio2023landmines,
  title        = {Test Images of Buried landmines},
  author       = {Tenorio Tamayo, Alejandro and
                  Forero Ramirez, Juan Camilo and
                  Garc{\'i}a, Bryan and
                  Loaiza Correa, Humberto and
                  Restrepo Gir{\'o}n, Andr{\'e}s David and
                  Nope, Sandra Esperanza and
                  Barandica L{\'o}pez, Asfur and
                  Buitrago, Jos{\'e} Tom{\'a}s},
  howpublished = {Mendeley Data, V4},
  year         = {2023},
  doi          = {10.17632/732ngnf4r3.4},
  url          = {https://data.mendeley.com/datasets/732ngnf4r3/4},
  note         = {Universidad del Valle. Licensed CC BY 4.0}
}

Authors: Alejandro Tenorio Tamayo, Juan Camilo Forero Ramirez, Bryan García, Humberto Loaiza Correa, Andrés David Restrepo Girón, Sandra Esperanza Nope, Asfur Barandica López, José Tomás Buitrago (Universidad del Valle).

License

CC BY 4.0, inherited from the original dataset. You may share and adapt the material, including commercially, provided you give appropriate credit, link the licence, and indicate any changes. The only change here is the archive format (.rar → .zip).