kaityc06/iNat21_masked

iNat21 Masked

Segmentation masks for the iNaturalist 2021 training set, stored as RLE-encoded masks in a single Parquet file.

File

File	Description
masks.parquet	173,548 rows — one per image — with RLE mask, score, species label, and file_id

Schema

Column	Type	Description
file_id	str	UUID stem of the original iNat21 image filename
dataset	str	Always iNat21
text_prompt	str	Text prompt used to generate the mask
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
image_width	int	Original image width
image_height	int	Original image height
supercategory	str	Broad taxonomic group (e.g. Plants, Birds)
species	str	Species name

Loading the masks

import pandas as pd

df = pd.read_parquet("hf://datasets/kaityc06/iNat21_masked/masks.parquet")

Retrieving the original image from a file_id

The file_id matches the filename stem of the corresponding iNat21 image (e.g. d615f184-8af4-4c60-b9f8-3081c1607644 → train/<category>/<file_id>.jpg).

Build a lookup once from the iNat21 train.json annotation file, then open any image by its file_id:

import json
import os
from PIL import Image

# Path to your local iNat21 cache (must contain train.json and train/)
INAT21_CACHE = "/path/to/inat21_cache"

# Build file_id → absolute path mapping once
with open(os.path.join(INAT21_CACHE, "train.json")) as f:
    data = json.load(f)

file_id_to_path = {
    os.path.basename(img["file_name"]).rsplit(".", 1)[0]: os.path.join(INAT21_CACHE, img["file_name"])
    for img in data["images"]
}

# Look up any file_id from the parquet
file_id = "d615f184-8af4-4c60-b9f8-3081c1607644"
img = Image.open(file_id_to_path[file_id])

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