🧪 SmellDB

The first ML benchmark for artificial olfaction — evaluations coming soon.

Overview

Smell is perhaps our most complicated and misunderstood sense. SmellDB is the first publicly available benchmark with a dataset that benchmarks AI models on olfactory data.

At the moment, entries are API-first. Download the dataset, build your model, and submit predictions via our REST API. Scores are returned instantly and performance is optionally logged on a public leaderboard.

The held-out test data as well as the benchmark will be made available in the near future.

Highlights

🗂 Samples	3,000 labelled samples
⏱ Duration	115 seconds per sample
📡 Channels	32 sensor channels
🕐 Total recording	96 hours
🏷 Classes	12 semantic smell labels

The Task

Given multichannel time-series sensor data from a 32-channel electronic nose (e-Nose), predict the 1-of-12 smell label for each sample.

#	Label
1	🍊 Sweet Orange
2	🪵 Leather & Tobacco
3	🍂 Cinnamon Leaf
4	🍌 Banana
5	☕ Cafe Latte
6	🌹 Rose
7	💜 Lavender Oil
8	🥥 Coconut
9	🥭 Mango
10	🌸 English Orchid
11	🧴 Monkey Farts
12	💧 Water

Getting Started

1. Get your API key

Register by making two API calls to https://smelldb.anemolabs.com.

( a ) - Register. A 6-digit verification code will be sent to your email.

curl -X POST https://smelldb.anemolabs.com/v1/user/register \
  -H "Content-Type: application/json" \
  -d '{"email": "you@example.com", "name": "Your Name", "organisation": "Optional"}'

( b ) - Activate. Your API key will be emailed to you on success.

curl -X POST https://smelldb.anemolabs.com/v1/user \
  -H "Content-Type: application/json" \
  -d '{"action": "activate_user", "email": "you@example.com", "verification_code": "123456"}'

( c ) - Rotate API key (optional). If you need to invalidate your current key, a new one will be emailed to you.

curl -X POST https://smelldb.anemolabs.com/v1/user \
-H "Content-Type: application/json" \`
-d '{"action": "request_new_api_key", "email": "you@example.com"}'

---

2. Download the dataset

https://smelldb.anemolabs.com/dataset/download

YOUR_DIRECTORY/
├── dataset.csv     # sensor readings (N samples × 32 channels) and labels (final 'LABEL' column)

---

3. Build your model

import pandas as pd

df = pd.read_csv("YOUR_DIRECTORY/dataset.csv")
y = df['LABEL']
X = df.drop('LABEL', axis=1)

# Transform Data

# Train model
model.fit(X,y)

# Generate smell label predictions here
predictions = model.predict(X)   # list of label strings

---

4. Submit

import requests

response = requests.post(
    "https://smelldb.anemolabs.com/v1/evals",
    headers={
        "X-API-Key": "YOUR_API_KEY",
        "Content-Type": "application/json",
    },
    json={
        "dataset_name": "smelldb-base-v1",
        "predictions": predictions.tolist(),
    },
)

print(response.json())

---

5. View your submissions

response = requests.get(
    "https://smelldb.anemolabs.com/v1/evals",
    headers={
        "X-API-Key": "YOUR_API_KEY",
    },
)
print(response.json())