File size: 1,816 Bytes
855a0e3
3082b8e
 
855a0e3
 
 
 
318a949
 
 
 
 
 
855a0e3
 
3082b8e
 
 
 
 
 
 
855a0e3
 
 
 
 
3082b8e
 
855a0e3
542a5a3
4625f0e
375ead1
4625f0e
 
 
 
 
2c0f883
4625f0e
2c0f883
4625f0e
 
 
 
 
 
 
0e4a518
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
---
size_categories:
- 10K<n<100K
dataset_info:
  features:
  - name: image
    dtype: image
  - name: left
    dtype: int64
  - name: forward
    dtype: int64
  - name: right
    dtype: int64
  splits:
  - name: train
    num_bytes: 561073580.031
    num_examples: 12489
  - name: test
    num_bytes: 60984278.384
    num_examples: 1388
  download_size: 618329781
  dataset_size: 622057858.415
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: test
    path: data/test-*
---
# Image Dataset of Cyberzoo at Delft University of Technology

This dataset includes images taken in the Cyberzoo in the aircraft hall of the Delft University of Technology. The dataset consists of both real images (82%) and simulator images (18%).

## Dataset Details

### Dataset Description

This dataset was collected throughout multiple testing sessions at the Cyberzoo, both while actually flying and handheld. The labeling of the data has been performed using monocular depth maps, generated using [Depth-Anything](https://github.com/LiheYoung/Depth-Anything). The exact labeling process has been explained in [this](https://github.com/Timdnb/CNN-for-Micro-Air-Vehicles/blob/main/Dataset_generation.ipynb) notebook.

- **Curated by:** [Tim den Blanken](https://github.com/Timdnb)

## Uses

This dataset can be used to train Convolutional Neural Networks for obstacle avoidance of Micro Air Vehicles in the Cyberzoo of Delft University of Technology. For the entire training pipeline, please go this [this](https://github.com/Timdnb/CNN-for-Micro-Air-Vehicles) repository.

## Dataset Structure

The dataset consists of a train set (90% of the data) and a test set (10% of the data). Each image has its label embedded in the metadata, the possible labels are: "left", "forward", "right".