Dataset of raw IQ recordings (RF signals from 37 UAV/drone types)

https://grok.com/share/c2hhcmQtNA_0c2f8c75-7a93-4883-9d8d-a13ebe5b66dd

Hey Grok, I have a Disk Space crunch, how could I use https://huggingface.co/datasets/kitofrank/RFUAV without downloading to whole 10GB?

Yes, you can use the kitofrank/RFUAV dataset without downloading the full ~10GB (or larger raw data).

The dataset is hosted on Hugging Face and consists of raw IQ recordings (RF signals from 37 UAV/drone types), spectrogram images, and some model weights/preprocessing files. The bulk of the size comes from the raw binary IQ data and image folders.

Best Options for Disk-Constrained Use (Ranked by Convenience)

1. Streaming Mode (Recommended – Zero Disk Usage)

The Hugging Face datasets library can stream the data on-the-fly. It downloads and processes one example at a time without saving the full dataset to disk.

from datasets import load_dataset

# Stream the dataset (no full download)
dataset = load_dataset("kitofrank/RFUAV", streaming=True)

# Example: iterate over train split (or specific config/split if available)
for example in dataset["train"]:
    # example contains IQ data, spectrogram, labels, etc.
    iq_samples = example["iq"]          # raw I/Q signal
    spectrogram = example["image"]      # PIL Image or array
    label = example["label"]            # drone type / class
    snr = example.get("snr")            # signal-to-noise ratio if present

    # Process only what you need (e.g., convert IQ to spectrogram on-the-fly)
    # Then feed into your Gemma-3n reasoning, embeddinggemma, or hypergraph
    print(f"Drone type: {label}, SNR: {snr}")

    # Stop early if you only want a subset
    if some_condition:
        break

Advantages:

Almost zero persistent disk usage (only temporary buffers).
Perfect for your RF pipeline — process one UAV signal at a time, extract features, embed with embeddinggemma, reason with Gemma-3n, then discard.
Works great with torch.utils.data.DataLoader for batching.

Tip: If the dataset has multiple splits/configs (e.g., train, valid, raw vs. spectrogram), specify them:

dataset = load_dataset("kitofrank/RFUAV", "default", streaming=True, split="train")

2. Download Only a Subset / Slice

If you need a small fixed portion on disk:

# First 500 examples only (downloads only what’s needed)
small_dataset = load_dataset("kitofrank/RFUAV", split="train[:500]")

# Or percentage (if supported for this dataset)
small_dataset = load_dataset("kitofrank/RFUAV", split="train[:5%]")

You can also specify data_files if you know the Parquet/ file structure, but streaming is usually simpler.

3. Download Only Specific Files or Columns

Use the Hugging Face Hub API or hf_hub_download to grab individual files (e.g., one drone’s raw data or spectrograms).
If the dataset has Parquet files, you can download only certain Parquet shards.

Example:

from huggingface_hub import hf_hub_download

# Download one specific file (replace with actual path from the repo tree)
hf_hub_download(repo_id="kitofrank/RFUAV", filename="weight/some_model_weights.bin", local_dir="rfuav_subset")

4. Hybrid Approach for Your Workflow

Since you’re working with RF fusion, hypergraph, Gemma-3n reasoning, and Three.js viz:

Stream most of the time for exploration and feature extraction.
Download small targeted subsets (e.g., only high-SNR samples from 5–10 drone types) when you need offline training data.
Use datasets map() or filter() with streaming to preprocess on-the-fly (e.g., convert raw IQ → spectrogram → embedding).

Practical Next Steps for Your RF Scythe Setup

Start with streaming + a small slice to prototype:

   ds = load_dataset("kitofrank/RFUAV", streaming=True, split="train[:1000]")

Feed examples into your pipeline:

Extract IQ or spectrogram → run local spectrogram analysis (your DASPy code).
Generate text description → embed with embeddinggemma.
Feed to Gemma-3n with one of the RF reasoning templates we discussed earlier.

Store only the processed features (embeddings, labels, metadata) in your local DuckDB/Parquet store — discard raw data immediately.

This way you get the value of the RFUAV benchmark (real drone RF signatures for detection/ID) without the 10GB+ disk hit.

Would you like a ready-to-run code snippet that:

Streams a few examples
Converts IQ to spectrogram
Embeds the description
Runs a Gemma-3n reasoning prompt
Saves only the compact results?

Just let me know your preferred output format or next integration point (e.g., hypergraph node creation).

kitofrank/RFUAV · Datasets at Hugging Face

This repository contains the RFUAV dataset , presented in the paper ” RFUAV : A Benchmark Dataset for Unmanned Aerial Vehicle Detection and Identification”.

Huggingface kitofrank/RFUAV · Datasets at Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science.Huggingface kitofrank/RFUAV at mainDatasets : · kitofrank . /. RFUAV . like 19 ; Tasks: Audio Classification ; Modalities: Image ; Formats: imagefolder ; Languages: English ; Size : 10K – 100K.Huggingface README.md · kitofrank/RFUAV at …The RFUAV DATASET . Abstract. The official repository for our paper, ” RFUAV : A Benchmark Dataset for Unmanned Aerial Vehicle Detection and …Huggingface kitofrank/RFUAV at mainTasks: Audio Classification ; Modalities: Image ; Formats: imagefolder ; Languages: English ; Size : 10K – 100K.Huggingface 961 BytesRFUAV offers a comprehensive benchmark dataset for Radio-Frequency (RF)-based drone detection and identification. ![pic.1](./abstract/profile.png) More detailed …Huggingface weight/config-template.yaml · kitofrank/RFUAV at mainDatasets : · kitofrank . /. RFUAV . like 20 ; Tasks: Audio Classification ; Modalities: Image ; Formats: imagefolder ; Languages: English ; Size : 10K – …Huggingface TasksDatasets : · kitofrank . /. RFUAV . like 19 ; Tasks: Audio Classification ; Modalities: Image ; Formats: imagefolder ; Languages: English ; Size : 10K – 100K.Huggingface RAW DATADatasets : · kitofrank . /. RFUAV . like 20 ; Tasks: Audio Classification ; Modalities: Image ; Formats: imagefolder ; Languages: English ; Size : 10K – …Huggingface kitoweeknd/RFUAV: This is official …The official repository for our paper, ” RFUAV : A Benchmark Dataset for Unmanned Aerial Vehicle Detection and Identification”, can be accessed here.Github

Searched webhow to stream or download subset of Hugging Face dataset without full download10

Is there is a way that I can download only a part …

Did you consider using the stream feature in Datasets ? It allows you to stream data from HuggingFace’s Hub without having to download the dataset locally.

Stackoverflow Download only a subset of a split – 🤗DatasetsIf your dataset is too big, please use streaming mode. You can also slice your dataset in streaming mode, see the documentation here: Stream .Discuss.huggingface How to get the number of samples in a dataset without …The only option then is to use the streaming feature to iterate over a dataset’s samples but without downloading it.Discuss.huggingface StreamDataset streaming lets you work with a dataset without downloading it . The data is streamed as you iterate over the dataset. This is especially helpful when:.Huggingface How To Download & Process Hugging Face DatasetLet’s see how we can download and process any hugging face data set . Be it on Google Collab or on your local machine the same code will work.Youtube How to process the first 20k samples of a dataset without …Just 20k samples per locale is enough for my purpose. But I cannot find a way to select only 20k first samples without downloading the whole dataset first.Discuss.huggingface Loading a fraction of data – 🤗DatasetsYou can load a subset from a dataset this way: subset = load_dataset (…, split=”train[:30%]”) Note that it still downloads and prepares the full dataset .Discuss.huggingface Allow downloading just some columns of a dataset #4114Hi @lhoestq, I have an audio dataset of 250GB on the huggingface hub in parquet format. I only wanted to load the text column. It is taking a …Github How can I download a sizable subset of a datasetHi, I would like to download 5% of the German dataset from allenai/c4. It is a huge, 300+ billion-word dataset , so its 5% is also very big.Discuss.huggingface Simplest Way to Download Models and Datasets from …This video shares couple of easiest tools to use to download the models and datasets from Hugging Face quickly. Buy Me a Coffee to support …Youtube

Searched webkitofrank RFUAV dataset description RF UAV signals5