Datasets:
File size: 4,783 Bytes
bf82c28 b5e3aba bf82c28 b5e3aba bf82c28 5ce23d0 bf82c28 b5e3aba 5a65331 d6f756b 5a65331 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 |
---
license: afl-3.0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
dataset_info:
features:
- name: image
dtype: image
- name: width
dtype: int64
- name: height
dtype: int64
- name: objects
struct:
- name: bbox
sequence:
sequence: int64
- name: categories
sequence: int64
- name: area
sequence: float64
splits:
- name: train
num_bytes: 1902369683.516
num_examples: 3924
download_size: 1670295565
dataset_size: 1902369683.516
task_categories:
- object-detection
tags:
- aerial-images
- savmap
- wildlife
- savanna
pretty_name: savmap-annotated
---
# Dataset Card for SAVMAP (Enhanced Annotations)
## Dataset Summary
The SAVMAP dataset comprises ultrahigh-resolution aerial imagery captured by unmanned aerial vehicles (UAVs) over semi-arid savanna landscapes in Namibia. Originally released in 2015, the dataset includes raw JPEG images with embedded EXIF metadata (timestamp, latitude, longitude, altitude) and annotations identifying animals in the imagery.
⚠️ **This version of the dataset includes improved annotations** to address issues found in the original Micromappers campaign, which was known to include many false positives. The updated annotations were curated and validated by **Paul Allin**, a PhD candidate at **Stellenbosch University (South Africa)**, providing a more accurate basis for training and evaluating computer vision models in ecological monitoring tasks.
⚠️ **Disclaimer**: this dataset is not identical to the "version 2" hosted on Zenodo.org. The images have been cropped to 2000*2000 and only a portion of the negative samples (i.e. empty images) have been selected. The dataset comprises 3545 negative samples and 379 positive samples.
## Supported Tasks and Use Cases
- **Object Detection**: Training and evaluation of animal detection models in ultrahigh-resolution aerial imagery.
- **Remote Sensing for Ecology**: Analyzing habitat use and wildlife distributions in savanna environments.
- **Biodiversity Conservation**: Informing sustainable land-use strategies and conservation interventions.
## Languages
Non-linguistic (imagery and geospatial data); metadata and documentation are provided in English.
## Dataset Creation
### Curation Rationale
The original SAVMAP dataset was developed to demonstrate the use of UAV imaging for near real-time ecological monitoring and sustainable land-use planning. However, the reliance on crowdsourced annotations introduced quality issues. To make the dataset more useful for machine learning and conservation efforts, a new set of expert-curated annotations has been developed.
### Source Data
- **Imagery**: Captured in May 2014 over a Namibian savanna reserve using EPFL-designed UAVs.
- **Expert Annotations**: Created by Paul Allin (Stellenbosch University), focusing on accuracy, removing false positives, and validating animal locations.
## Annotation Process
- **Enhanced (Paul Allin)**: Bounding boxes were created and reviewed manually by an ecologist with domain expertise, ensuring significantly improved data quality. The dataset has 379 positive samples (image with annotations) and 3545 negative samples (images without annotations).
## Licensing Information
The dataset is released under the **Academic Free License v3.0**. Please cite the original dataset creators and acknowledge the enhanced annotations as described below.
## Citation Information
Please cite both the original creators and the updated annotation contributor as follows:
**Original Dataset**:
> Reinhard, F., Parkan, M., Produit, T., Betschart, S., Bacchilega, B., Hauptfleisch, M. L., Meier, P., SAVMAP Consortium, & Joost, S. (2015). Near real-time ultrahigh-resolution imaging from unmanned aerial vehicles for sustainable land use management and biodiversity conservation in semi-arid savanna under regional and global change (SAVMAP) (Version 2.0). Zenodo. https://doi.org/10.5281/zenodo.1204408
**Enhanced Annotations**:
> Enhanced annotations curated by Paul Allin, PhD candidate, Stellenbosch University, South Africa (2025). For correspondence regarding expert annotations, contact fadel.seydou@gmail.com.
## Contributions
- **Original Dataset Team**:
Friedrich Reinhard, Matthew Parkan, Timothée Produit, Sonja Betschart, Morgan L. Hauptfleisch, Patrick Meier, Beatrice Bacchilega, Stéphane Joost, SAVMAP Consortium.
- **Enhanced Annotations**:
Paul Allin, PhD candidate, Stellenbosch University, South Africa.
- **Dataset Engineering and Integration**:
Fadel Mamar Seydou, Independent ML researcher – contributed to dataset cleanup, reformatting annotations for ML pipelines, and preparing the release for Hugging Face compatibility.
|