kaityc06/iWildCam_masked

iWildCam Masked

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

File

File	Description
masks.parquet	30,959 rows — one per image — with RLE mask, score, species label, location, and split

Schema

Column	Type	Description
dataset	str	Always iWildCam
text_prompt	str	Visual concept term used to generate the mask (e.g. tiger, elephant, bird)
mask_rle_counts	str	RLE-encoded mask (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 (range 0.30–0.99, mean 0.88)
image_width	int	Original image width
image_height	int	Original image height
species	str	Scientific name of the species (e.g. panthera tigris)
location	int	Camera trap location ID from the iWildCam metadata
split	str	Dataset split: train, id_val, id_test, or test

Split breakdown

Split	Rows
train	14,639
test	13,925
id_test	1,229
id_val	1,166
Total	30,959

Coverage: 108 species across 219 camera trap locations. Empty-class images (y == 0) are excluded.

Loading the masks

import pandas as pd

df = pd.read_parquet("hf://datasets/kaityc06/iWildCam_masked/masks.parquet")

Retrieving the original image

Images are stored as UUID-named JPEG files under the train/ directory. Cross-reference the species column against data/categories.csv and look up the filename in data/metadata.csv:

import os
import pandas as pd
from PIL import Image

# Path to your local iWildCam data directory (must contain metadata.csv and train/)
IWILDCAM_DATA = "/path/to/iWildCam/data"

# Build a lookup from (species, location, split) → list of image filenames
meta = pd.read_csv(os.path.join(IWILDCAM_DATA, "metadata.csv"))
cats = pd.read_csv(os.path.join(IWILDCAM_DATA, "categories.csv"))
label_to_species = dict(zip(cats["y"], cats["name"]))
meta["species"] = meta["y"].map(label_to_species)

# Index by species for quick lookup
species_to_rows = meta.groupby("species")

# Given a row from masks.parquet, find its image
df = pd.read_parquet("masks.parquet")
row = df.iloc[0]

candidates = species_to_rows.get_group(row["species"])
# Optionally filter by location and split to narrow to exact image
candidates = candidates[
    (candidates["location"] == row["location"]) &
    (candidates["split"] == row["split"])
]
img_path = os.path.join(IWILDCAM_DATA, "train", candidates.iloc[0]["filename"])
img = Image.open(img_path)

Decoding a mask

import numpy as np
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)  # numpy array, shape (H, W), dtype uint8