lance-format/stanford-cars-lance

Stanford Cars (Lance Format)

A Lance-formatted version of the Stanford Cars fine-grained benchmark — 8,144 photographs across 196 make/model/year classes — sourced from Multimodal-Fatima/StanfordCars_train. Each row carries the inline JPEG bytes, the integer class id, a BLIP-generated caption inherited from the source mirror, and a cosine-normalized CLIP image embedding, all available directly from the Hub at hf://datasets/lance-format/stanford-cars-lance/data.

Key features

• Inline JPEG bytes in the image column — no sidecar files, no image folders.
• Pre-computed CLIP image embeddings (image_emb, OpenCLIP ViT-B-32, 512-dim, cosine-normalized) with a bundled IVF_PQ index for similarity search.
• BLIP captions in blip_caption with a full-text index, so keyword search on visual descriptions composes with vector search in a single query.
• A bundled scalar index on label makes class-based curation a cheap predicate rather than a full scan.

Splits

Split	Rows	Notes
train.lance	8,144	The source mirror redistributes a single split; the original Stanford Cars test split is not included here.

Schema

Column	Type	Notes
id	int64	Row index within split (natural join key for merges)
image	large_binary	Inline JPEG bytes (quality 92)
label	int32	Class id (0–195), one per Make Model Year combination
blip_caption	string?	BLIP-generated caption (beam=5) carried through from the source mirror
image_emb	fixed_size_list<float32, 512>	OpenCLIP ViT-B-32 image embedding (cosine-normalized)

Pre-built indices

• IVF_PQ on image_emb — vector similarity search (cosine)
• INVERTED (FTS) on blip_caption — keyword and hybrid search
• BTREE on label — fast lookup by class id

Why Lance?

1. Blazing Fast Random Access: Optimized for fetching scattered rows, making it ideal for random sampling, real-time ML serving, and interactive applications without performance degradation.
2. Native Multimodal Support: Store text, embeddings, and other data types together in a single file. Large binary objects are loaded lazily, and vectors are optimized for fast similarity search.
3. Native Index Support: Lance comes with fast, on-disk, scalable vector and FTS indexes that sit right alongside the dataset on the Hub, so you can share not only your data but also your embeddings and indexes without your users needing to recompute them.
4. Efficient Data Evolution: Add new columns and backfill data without rewriting the entire dataset. This is perfect for evolving ML features, adding new embeddings, or introducing moderation tags over time.
5. Versatile Querying: Supports combining vector similarity search, full-text search, and SQL-style filtering in a single query, accelerated by on-disk indexes.
6. Data Versioning: Every mutation commits a new version; previous versions remain intact on disk. Tags pin a snapshot by name, so retrieval systems and training runs can reproduce against an exact slice of history.

Load with datasets.load_dataset

You can load Lance datasets via the standard HuggingFace datasets interface, suitable when your pipeline already speaks Dataset / IterableDataset or you want a quick streaming sample without installing anything Lance-specific.

import datasets

hf_ds = datasets.load_dataset("lance-format/stanford-cars-lance", split="train", streaming=True)
for row in hf_ds.take(3):
    print(row["label"], row["blip_caption"])

Load with LanceDB

LanceDB is the embedded retrieval library built on top of the Lance format (docs), and is the interface most users interact with. It wraps the dataset as a queryable table with search and filter builders, and is the entry point used by the Search, Curate, Evolve, Train, Versioning, and Materialize-a-subset sections below.

import lancedb

db = lancedb.connect("hf://datasets/lance-format/stanford-cars-lance/data")
tbl = db.open_table("train")
print(len(tbl))

Load with Lance

pylance is the Python binding for the Lance format and works directly with the format's lower-level APIs. Reach for it when you want to inspect or operate on dataset internals — schema, scanner, fragments, and the list of pre-built indices.

import lance

ds = lance.dataset("hf://datasets/lance-format/stanford-cars-lance/data/train.lance")
print(ds.count_rows(), ds.schema.names)
print(ds.list_indices())

Tip — for production use, download locally first. Streaming from the Hub works for exploration, but heavy random access and ANN search are far faster against a local copy:

hf download lance-format/stanford-cars-lance --repo-type dataset --local-dir ./stanford-cars-lance

Then point Lance or LanceDB at ./stanford-cars-lance/data.

Search

The bundled IVF_PQ index on image_emb makes approximate-nearest-neighbor search a single call. In production you would encode a query photo through the same OpenCLIP ViT-B-32 model used at ingest and pass the resulting 512-d vector to tbl.search(...). The example below uses the embedding stored in row 0 as a runnable stand-in so the snippet works without any model loaded.

import lancedb

db = lancedb.connect("hf://datasets/lance-format/stanford-cars-lance/data")
tbl = db.open_table("train")

seed = (
    tbl.search()
    .select(["image_emb", "blip_caption"])
    .limit(1)
    .to_list()[0]
)

hits = (
    tbl.search(seed["image_emb"])
    .metric("cosine")
    .select(["id", "label", "blip_caption"])
    .limit(10)
    .to_list()
)
print("seed caption:", seed["blip_caption"])
for r in hits:
    print(f"  {r['id']:>6}  label={r['label']:>3}  {r['blip_caption'][:60]}")

Tune metric, nprobes, and refine_factor to trade recall against latency for your workload.

Because the dataset also ships an INVERTED index on blip_caption, the same query can be issued as a hybrid search that combines the dense vector with a keyword query. LanceDB merges the two result lists and reranks them in a single call, which is useful when a phrase like "red convertible" must literally appear in the caption but you still want CLIP to do the heavy lifting on visual similarity.

hybrid_hits = (
    tbl.search(query_type="hybrid")
    .vector(seed["image_emb"])
    .text("red convertible")
    .select(["id", "label", "blip_caption"])
    .limit(10)
    .to_list()
)
for r in hybrid_hits:
    print(f"  {r['id']:>6}  label={r['label']:>3}  {r['blip_caption'][:60]}")

Curate

A typical curation pass for a fine-grained classifier combines a class-based filter with a content filter on the caption. Stacking both inside a single filtered scan keeps the result small and explicit, and the bounded .limit(200) makes it cheap to inspect before committing the subset to anything downstream. The BTREE on label and the INVERTED index on blip_caption make both predicates effectively free.

import lancedb

db = lancedb.connect("hf://datasets/lance-format/stanford-cars-lance/data")
tbl = db.open_table("train")

candidates = (
    tbl.search("convertible OR coupe")
    .where("label IN (12, 47, 89)", prefilter=True)
    .select(["id", "label", "blip_caption"])
    .limit(200)
    .to_list()
)
print(f"{len(candidates)} candidates; first: {candidates[0]['blip_caption'][:80]}")

The result is a plain list of dictionaries, ready to inspect, persist as a manifest of ids, or feed into the Evolve and Train workflows below. The image column is never read, so the network traffic for a 200-row candidate scan is dominated by the small caption strings rather than JPEG bytes.

Evolve

Lance stores each column independently, so a new column can be appended without rewriting the existing data. The lightest form is a SQL expression: derive the new column from columns that already exist, and Lance computes it once and persists it. Stanford Cars class strings often encode the model year as a trailing four-digit token in the caption; the example below uses a SQL regex to lift that year into its own column, and adds a flag for vintage cars. Either can then be used directly in where clauses without recomputing the predicate on every query.

Note: Mutations require a local copy of the dataset, since the Hub mount is read-only. See the Materialize-a-subset section at the end of this card for a streaming pattern that downloads only the rows and columns you need, or use hf download to pull the full split first.

import lancedb

db = lancedb.connect("./stanford-cars-lance/data")  # local copy required for writes
tbl = db.open_table("train")

tbl.add_columns({
    "caption_year": "CAST(regexp_extract(blip_caption, '(\\d{4})', 1) AS INTEGER)",
    "is_long_caption": "length(blip_caption) >= 80",
})

If the values you want to attach already live in another table (offline labels, classifier predictions, the Make Model Year strings from the original Stanford metadata), merge them in by joining on id:

import pyarrow as pa

class_strings = pa.table({
    "id": pa.array([0, 1, 2]),
    "class_name": pa.array([
        "AM General Hummer SUV 2000",
        "Acura RL Sedan 2012",
        "Acura TL Sedan 2012",
    ]),
})
tbl.merge(class_strings, on="id")

The original columns and indices are untouched, so existing code that does not reference the new columns continues to work unchanged. New columns become visible to every reader as soon as the operation commits. For column values that require a Python computation (e.g., running a second captioner over the image bytes), Lance provides a batch-UDF API in the underlying library — see the Lance data evolution docs for that pattern.

Train

Projection lets a training loop read only the columns each step actually needs. LanceDB tables expose this through Permutation.identity(tbl).select_columns([...]), which plugs straight into the standard torch.utils.data.DataLoader so prefetch, shuffling, and batching behave as in any PyTorch pipeline. For a from-scratch fine-grained classifier, project the JPEG bytes and the integer label; for a linear probe on top of frozen CLIP features, swap the projection to the embedding column and skip JPEG decoding entirely.

import lancedb
from lancedb.permutation import Permutation
from torch.utils.data import DataLoader

db = lancedb.connect("hf://datasets/lance-format/stanford-cars-lance/data")
tbl = db.open_table("train")

train_ds = Permutation.identity(tbl).select_columns(["image", "label"])
loader = DataLoader(train_ds, batch_size=64, shuffle=True, num_workers=4)

for batch in loader:
    # batch carries only the projected columns; image_emb and blip_caption stay on disk.
    # decode the JPEG bytes, forward, cross-entropy against `label`...
    ...

Switching feature sets is a configuration change: passing ["image_emb", "label"] to select_columns(...) on the next run reads only the cached 512-d vectors and the label, which is the right shape for a linear probe or a lightweight reranker.

Versioning

Every mutation to a Lance dataset, whether it adds a column, merges labels, or builds an index, commits a new version. Previous versions remain intact on disk. You can list versions and inspect the history directly from the Hub copy; creating new tags requires a local copy since tags are writes.

import lancedb

db = lancedb.connect("hf://datasets/lance-format/stanford-cars-lance/data")
tbl = db.open_table("train")

print("Current version:", tbl.version)
print("History:", tbl.list_versions())
print("Tags:", tbl.tags.list())

Once you have a local copy, tag a version for reproducibility:

local_db = lancedb.connect("./stanford-cars-lance/data")
local_tbl = local_db.open_table("train")
local_tbl.tags.create("clip-vitb32-v1", local_tbl.version)

A tagged version can be opened by name, or any version reopened by its number, against either the Hub copy or a local one:

tbl_v1 = db.open_table("train", version="clip-vitb32-v1")
tbl_v5 = db.open_table("train", version=5)

Pinning supports two workflows. A retrieval system locked to clip-vitb32-v1 keeps returning stable results while the dataset evolves in parallel; newly added columns or captions do not change what the tag resolves to. A training experiment pinned to the same tag can be rerun later against the exact same images and labels, so changes in metrics reflect model changes rather than data drift. Neither workflow needs shadow copies or external manifest tracking.

Materialize a subset

Reads from the Hub are lazy, so exploratory queries only transfer the columns and row groups they touch. Mutating operations (Evolve, tag creation) need a writable backing store, and a training loop benefits from a local copy with fast random access. Both can be served by a subset of the dataset rather than the full split. The pattern is to stream a filtered query through .to_batches() into a new local table; only the projected columns and matching row groups cross the wire, and the bytes never fully materialize in Python memory.

import lancedb

remote_db = lancedb.connect("hf://datasets/lance-format/stanford-cars-lance/data")
remote_tbl = remote_db.open_table("train")

batches = (
    remote_tbl.search("convertible OR coupe")
    .select(["id", "image", "label", "blip_caption", "image_emb"])
    .to_batches()
)

local_db = lancedb.connect("./stanford-cars-sports-subset")
local_db.create_table("train", batches)

The resulting ./stanford-cars-sports-subset is a first-class LanceDB database. Every snippet in the Evolve, Train, and Versioning sections above works against it by swapping hf://datasets/lance-format/stanford-cars-lance/data for ./stanford-cars-sports-subset.

Source & license

Converted from Multimodal-Fatima/StanfordCars_train, itself a redistribution of the Stanford Cars dataset. The original dataset license is for non-commercial research use; review the Stanford Cars terms before redistribution.

Citation

@inproceedings{krause2013collecting,
  title={Collecting a large-scale dataset of fine-grained cars},
  author={Krause, Jonathan and Stark, Michael and Deng, Jia and Fei-Fei, Li},
  booktitle={Workshop on Fine-Grained Visual Categorization (CVPR)},
  year={2013}
}