Browse Datasets

Snapshot Serengeti

Camera-Trap Wildlife Images from Tanzania

image-classification object-detection ecological-monitoring ~5TB full (use a subset) CDLA-Permissive-1.0
Access Dataset

Description

Camera-trap image series from Serengeti National Park, Tanzania, for species classification and ecological monitoring; roughly three-quarters of frames are empty, creating strong class imbalance. The full archive is terabyte-scale, so use the provided metadata CSVs or a sampled subset for Colab.

Quick Start

Python 
# Browse metadata + download a subset from LILA (CDLA-Permissive):
#   https://lila.science/datasets/snapshot-serengeti/
import pandas as pd
# annotations = pd.read_json("SnapshotSerengeti_v2.json")

Project Ideas

Streaming Species Classifier

Stream a sampled subset into Colab to train a species classifier.

Advanced

Imbalance with Focal Loss

Tackle the ~76% empty-frame imbalance using focal loss.

Advanced

Ethical Considerations

  • Human images must be removed for privacy; discuss the surveillance implications of camera traps.