README.md

metadata

license: apache-2.0
task_categories:
  - image-segmentation
  - image-classification
language:
  - en
tags:
  - stanford-cars
  - segmentation
  - sam
  - bounding-box
  - mask
size_categories:
  - 10K<n<100K

Stanford Cars with SAM3 Segmentation Masks

This dataset was created using segmentationAPI.com

A version of the Stanford Cars dataset augmented with per-image segmentation masks generated by SAM3 (Segment Anything Model 3).

Dataset Description

• Images: 196 car model classes, original Stanford Cars images
• Masks: Binary segmentation masks from SAM3, stored as grayscale PNG (255 = car, 0 = background)
• Bounding boxes: [x1, y1, x2, y2] pixel coordinates of the primary car object
• Splits: train (8,144 images), test (8,041 images)

Schema

Column	Type	Description
image	Image	Original car photograph (JPEG)
label	int	Class label (0–195, 196 car models)
split	string	train or test
mask	Image	Binary segmentation mask (grayscale PNG)
bbox	list[int]	Bounding box [x1, y1, x2, y2] in pixels

Usage

from datasets import load_dataset

ds = load_dataset("segmentationAPIs/standford_cars_masks")
sample = ds["train"][0]

sample["image"]   # PIL Image — original photo
sample["mask"]    # PIL Image — grayscale mask (255 = car)
sample["bbox"]    # [x1, y1, x2, y2]
sample["label"]   # int class label

Class Labels

196 fine-grained car model classes covering major manufacturers (Acura, Aston Martin, Audi, BMW, Bentley, Bugatti, Buick, Cadillac, Chevrolet, Chrysler, Dodge, Ferrari, FIAT, Ford, GMC, Honda, Hyundai, Infiniti, Isuzu, Jaguar, Jeep, Lamborghini, Land Rover, Lincoln, MINI, Maybach, McLaren, Mercedes-Benz, Mitsubishi, Nissan, Plymouth, Pontiac, Porsche, Ram, Rolls-Royce, Scion, Spyker, Subaru, Suzuki, Tesla, Toyota, Volkswagen, Volvo).

Source

• Original dataset: Stanford Cars Dataset
• Masks generated with SAM3 (Segment Anything Model 3)