Hugging Face's logo

Datasets:
marco-willi
/
ser_sampled

Dataset card Files
xet
 Community
Dataset Viewer
The dataset viewer is not available for this subset.
Cannot get the split names for the config 'default' of the dataset.
Exception:    SplitsNotFoundError
Message:      The split names could not be parsed from the dataset config.
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.12/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
                  for split_generator in builder._split_generators(
                                         ^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 82, in _split_generators
                  raise ValueError(
              ValueError: The TAR archives of the dataset should be in WebDataset format, but the files in the archive don't share the same prefix or the same types.
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 65, in compute_split_names_from_streaming_response
                  for split in get_dataset_split_names(
                               ^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
                  info = get_dataset_config_info(
                         ^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
                  raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
              datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

license: cdla-permissive-1.0 task_categories:

  • image-classification tags:
  • camera-trap
  • wildlife
  • serengeti
  • snapshot-safari
  • megadetector pretty_name: Snapshot Safari SER Sampled — Classroom Subset v1.0 size_categories:
  • 1K<n<10K

Snapshot Safari SER (Serengeti) — Sampled Classroom Subset v1.0

Summary

A curated, realistically distributed subset of the Snapshot Safari 2024 Expansion SER (Serengeti National Park) camera trap dataset, prepared for use in the CAS Deep Learning — Computer Vision course exercises. Unlike ser_balanced, the class distribution reflects real-world Serengeti encounter rates: wildebeest and zebra dominate, while rarer species (impala, warthog, gazelle grants) appear infrequently.

Archive Images Description
ser_sampled.tar.gz 4 999 Realistic distribution, full frames
ser_sampled_cropped.tar.gz 4 949 Realistic distribution, MD-cropped

Source

MegaDetector

Pre-computed MegaDetector v1000-redwood RDE-filtered results from LILA Science: snapshot-safari-2024-expansion-SER-subset-v1000.0.0-redwood_detections.threshold.filtered.json.zip

Used to filter frames (conf ≥ 0.8) and select the best frame per sequence. The _cropped variant additionally crops each image to the primary detected animal bounding box (10% padding).

Species

buffalo, elephant, empty, gazellegrants, gazellethomsons, hartebeest, impala, warthog, wildebeestblue, zebraplains

Statistics — ser_sampled

Class Train Val Test Total
wildebeestblue 1 244 267 266 1 777
zebraplains 919 197 197 1 313
gazellethomsons 724 155 156 1 035
buffalo 146 31 31 208
elephant 115 25 25 165
hartebeest 115 25 24 164
gazellegrants 74 16 16 106
warthog 66 14 15 95
impala 60 13 13 86
empty 35 8 7 50
Total 3 498 751 750 4 999

Statistics — ser_sampled_cropped

Same splits and class labels as ser_sampled. The empty class is excluded from the cropped variant (no animal detection box available). Total: 4 949 images.

Note: ~56% of images are IR/night (near-infrared, nearly greyscale).

Curation Details

  • Deduplication: one image per sequence (highest MD animal confidence frame)
  • Animal filter: MD animal confidence ≥ 0.8
  • Empty filter: max MD animal confidence < 0.2
  • Sampling: proportional to real Serengeti encounter rates (no per-class cap)
  • Split strategy: stratified 70/15/15 by sequence ID — no sequence spans splits
  • Image resolution: resized to max 1024 px on longer side, JPEG quality 92
  • Format: ImageFolder layout — <split>/<label>/<filename>.jpg

Usage 

from huggingface_hub import hf_hub_download
import tarfile

# Full-frame version
archive = hf_hub_download(
    "marco-willi/ser_sampled",
    "ser_sampled.tar.gz",
    repo_type="dataset",
)
with tarfile.open(archive) as tar:
    tar.extractall(DATA_PATH)
# → DATA_PATH/ser_sampled/{train,val,test}/<label>/*.jpg

# MD-cropped version (recommended for Colab — smaller download)
archive = hf_hub_download(
    "marco-willi/ser_sampled_cropped",
    "ser_sampled_cropped.tar.gz",
    repo_type="dataset",
)
with tarfile.open(archive) as tar:
    tar.extractall(DATA_PATH)
# → DATA_PATH/ser_sampled_cropped/{train,val,test}/<label>/*.jpg
Downloads last month
14
Total file size:
1.22 GB