suryadv/Snapshot-Serengeti_masked

Snapshot Serengeti Masked

Segmentation masks for the Snapshot Serengeti dataset, stored as RLE-encoded masks in a single Parquet file.

File

File	Description
masks.parquet	53,487 rows — one per accepted mask — with RLE mask, score, source image metadata, prompt class, raw category label, and normalized bounding box when available

Schema

Column	Type	Description
dataset	str	Always SnapshotSerengeti
text_prompt	str	Text prompt used to generate the mask, e.g. zebra, wildebeest, bird, thomson gazelle
mask_rle_counts	str	RLE-encoded mask counts in COCO format
mask_rle_height	int	Height used for RLE decoding
mask_rle_width	int	Width used for RLE decoding
mask_score	float	Mask confidence score
image_width	int	Original image width
image_height	int	Original image height
image_id	str	Snapshot Serengeti image identifier
file_name	str	Original Snapshot Serengeti image path relative to the LILA image root
category	str	Raw Snapshot Serengeti category label
prompt_class	str	Normalized class prompt used for segmentation
bbox	str	Bounding box [x0, y0, x1, y1] normalized to [0, 1], or null/empty if unavailable

Loading the dataset

Load the Parquet file directly from Hugging Face.

import pandas as pd

df = pd.read_parquet(
    "hf://datasets/suryadv/Snapshot-Serengeti_masked/masks.parquet"
)

print(len(df))
print(df.columns.tolist())

Retrieving the original image

The file_name column stores the source image path under the Snapshot Serengeti LILA image root.

from io import BytesIO
from urllib.parse import quote

import requests
from PIL import Image

IMAGE_BASE_URL = (
    "https://lilawildlife.blob.core.windows.net/lila-wildlife/"
    "snapshotserengeti-unzipped/"
)

row = df.iloc[0]

file_name = row["file_name"].replace("\\", "/")
url = IMAGE_BASE_URL + quote(file_name, safe="/")

img = Image.open(
    BytesIO(requests.get(url, timeout=30).content)
).convert("RGB")

img

Retrieving an image by category

Look up an example image corresponding to a particular Snapshot Serengeti category.

from io import BytesIO
from urllib.parse import quote

import requests
from PIL import Image

IMAGE_BASE_URL = (
    "https://lilawildlife.blob.core.windows.net/lila-wildlife/"
    "snapshotserengeti-unzipped/"
)

category = "zebra"

row = df[
    df["category"].str.lower() == category.lower()
].iloc[0]

file_name = row["file_name"].replace("\\", "/")
url = IMAGE_BASE_URL + quote(file_name, safe="/")

img = Image.open(
    BytesIO(requests.get(url, timeout=30).content)
).convert("RGB")

img

Decoding a mask

Decode a COCO-format RLE mask into a binary NumPy array.

from pycocotools import mask as mask_utils

row = df.iloc[0]

rle = {
    "counts": row["mask_rle_counts"],
    "size": [
        row["mask_rle_height"],
        row["mask_rle_width"],
    ],
}

binary_mask = mask_utils.decode(rle)

print(binary_mask.shape)

The decoded mask is a NumPy array with shape (H, W) and dtype uint8.

License

This dataset is released under CC-BY-4.0.

Please also consult the original Snapshot Serengeti / LILA dataset terms when using the source imagery or metadata.