blackbeedrones/imav-2025-gate-dataset

IMAV 2025 Gate Detection Dataset

Object detection dataset for gate detection in the IMAV 2025 Indoor Competition - Mission 1. The MAV must navigate through tunnels formed by gates of three sizes (blue 1.5m, green 1.0m, red 0.5m) in a 10m x 10m indoor arena.

Trained model: blackbeedrones/imav-2025-gate

Competition Context

The 16th International Micro Air Vehicle Conference and Competition (IMAV 2025) took place in San Andres Cholula, Puebla, Mexico. The competition theme was "Search and Rescue", inspired by Mexico's seismic activity and the need for micro air vehicles in disaster response scenarios.

Mission 1: Enter the Room

The MAV must navigate through one of four entry options:

1. Free passage (0 pts): 1m wide open path
2. Wide tunnel (1 pt): 5 aligned gates, 1.5m x 1.5m window
3. Medium tunnel (2 pts): 5 aligned gates, 1.0m x 1.0m window
4. Small tunnel (3 pts): 5 aligned gates, 0.5m x 0.5m window

Target Object

• Tube diameter: 38.1 mm (1.5")
• Total height: 2m (base + window)
• Three tunnel sizes: Wide (blue, 1.5m), Medium (green, 1.0m), Small (red, 0.5m)

Dataset Structure

Split	Images
train	4081
validation	199
test	249

Total images: 4529

Classes: gate

Annotation format: COCO (x_min, y_min, width, height)

Usage

Load with HuggingFace Datasets

from datasets import load_dataset

dataset = load_dataset("blackbeedrones/imav-2025-gate-dataset")
example = dataset["train"][0]
print(example["objects"])  # {'bbox': [...], 'category': [...]}

Visualize with bounding boxes

import torch
from torchvision.ops import box_convert
from torchvision.utils import draw_bounding_boxes
from torchvision.transforms.functional import pil_to_tensor, to_pil_image

example = dataset["train"][0]
categories = dataset["train"].features["objects"].feature["category"]

boxes_xywh = torch.tensor(example["objects"]["bbox"])
boxes_xyxy = box_convert(boxes_xywh, "xywh", "xyxy")
labels = [categories.int2str(x) for x in example["objects"]["category"]]

to_pil_image(
    draw_bounding_boxes(
        pil_to_tensor(example["image"]),
        boxes_xyxy,
        colors="red",
        labels=labels,
    )
)

Convert to COCO JSON (for training)

import json
from datasets import load_dataset

dataset = load_dataset("blackbeedrones/imav-2025-gate-dataset", split="train")
categories = dataset.features["objects"].feature["category"]

coco = {
    "images": [],
    "annotations": [],
    "categories": [
        {"id": i, "name": n} for i, n in enumerate(categories.names)
    ],
}

ann_id = 0
for row in dataset:
    coco["images"].append({
        "id": row["image_id"],
        "width": row["width"],
        "height": row["height"],
        "file_name": f"{row['image_id']}.jpg",
    })
    row["image"].save(f"images/{row['image_id']}.jpg")

    for bbox, cat, area in zip(
        row["objects"]["bbox"],
        row["objects"]["category"],
        row["objects"]["area"],
    ):
        coco["annotations"].append({
            "id": ann_id,
            "image_id": row["image_id"],
            "category_id": cat,
            "bbox": bbox,
            "area": area,
            "iscrowd": 0,
        })
        ann_id += 1

with open("annotations.json", "w") as f:
    json.dump(coco, f)

Train with Nectar SDK

from nectar.ai.detection import Detector, TrainingConfig

detector = Detector("yolo11n.pt")
detector.load()
result = detector.train(TrainingConfig(
    dataset_path="path/to/converted/dataset",
    epochs=100,
    push_to_hub=True,
    hub_model_id="blackbeedrones/imav-2025-gate",
))

Train with Ultralytics (YOLO format)

from datasets import load_dataset
from pathlib import Path

dataset = load_dataset("blackbeedrones/imav-2025-gate-dataset")
categories = dataset["train"].features["objects"].feature["category"]

for split_name, split_data in dataset.items():
    img_dir = Path(f"yolo_dataset/images/{split_name}")
    lbl_dir = Path(f"yolo_dataset/labels/{split_name}")
    img_dir.mkdir(parents=True, exist_ok=True)
    lbl_dir.mkdir(parents=True, exist_ok=True)

    for row in split_data:
        fname = f"{row['image_id']}"
        row["image"].save(img_dir / f"{fname}.jpg")
        w, h = row["width"], row["height"]

        with open(lbl_dir / f"{fname}.txt", "w") as f:
            for bbox, cat in zip(row["objects"]["bbox"], row["objects"]["category"]):
                x_center = (bbox[0] + bbox[2] / 2) / w
                y_center = (bbox[1] + bbox[3] / 2) / h
                bw, bh = bbox[2] / w, bbox[3] / h
                f.write(f"{cat} {x_center} {y_center} {bw} {bh}\n")

References

• IMAV 2025
• Rulebook
• Nectar SDK
• Trained Model