This repo contains datasets that can be used for teaching. At the moment, it contains images/datasets mostly used for Computer Vision lectures.
The thermal-rgb folder and corresponding zip-files contain image pairs of thermal and RGB images recorded by a drone.
You can use the scenes ellipse, FH3, forest, and hut and are well suited for various tasks:
ellipse⬇️ (direct download): scene showing the FH's ellipse. Drone was hovering and only rotating around the up-axis. Perfect for panorama stitching.FH3⬇️ (direct download): scene showing the FH's FH3 building. Drone was flying at a horizontal distance of around 60m to the building. Suitable for panorama stitching.forest⬇️ (direct download): scene showing a forest. Drone was flying 50m above the ground. There is no overlap between the images, so panorama stitching is not possible.hut⬇️ (direct download): scene showing a hut in a zoo. Drone was flying in circular movements around the front of the hut. Might be suitable for panorama stitching (but will be hard).
An exemplary pair of the ellipse scene is shown in the following table.
| Thermal | RGB |
|---|---|
 |
 |
The hybrid_images folder and zip-file contains data for creating hybrid images.
The dataset is taken from a computer vision course at Georgia Tech and features pairs of aligned images.
An exemplary pair is shown in the following table.
| Image A | Image B |
|---|---|
 |
 |
⬇️ To directly download a zip folder of the dataset use this link!
The binary_leaves folder and .zip-file contais 282 binary images of 5 different leave types.
The binary images are taken from the Flavia leave dataset and feature a resolution of
| Japanese maple | Chinese cinnamon | ginkgo, maidenhair tree | Chinese tulip tree | tangerine |
|---|---|---|---|---|
 |
 |
 |
 |
 |
| 56 images | 55 images | 62 images | 53 images | 56 images |
⬇️ To directly download a zip folder of the dataset use this link!
The panorama_stitching folder and .zip-file contains images that can be used for panorama stitching algorithms.
Various images from different sources are included.
| Image(s) | Description, Source (Copyright) |
|---|---|
 (1250 x 442) |
This image is a panorama image of the campus Hagenberg showing the ellipse, FH1 and FH2. The image can be, for example, cropped in multiple regions and than stitched together. Source: Originally the image was hosted at the Website of the University. Sadly it is offline, now. This version was downloaded forman alternative: talente-ooe.at |
  (1024x683) |
Two images (A and B) recorded at theUniversity of Texas at Austin. The big tower in image B shows the Main Building. Source: the images are downloaded from the staff website of Hélio Pedrini. |
The example_images folder contains exemplary images that can be used for various tasks.
| Image | Description, Source |
|---|---|
 |
A cute cat image. Source:Wikimedia Commons |
 |
An image of a painting by Vincent van Gogh. Source:Wikimedia Commons |
 |
An image of Albert Einstein. Source:The Laboratory for Computational Vision is an interdisciplinary research group at New York University |
 |
An image of a woman. Useful for showcasing edge and line detection. Source:Flickr |
 |
An image of a sudoku puzzle. Useful for thresholding algorithms. Source:OpenCV |
The object_detection folder contains a small set of multi-object scenes used to demo modern detectors (YOLO, RT-DETR) and Vision-Language Models (e.g. Gemma 4). The classes shown — person, car, bus, apple, banana, orange — are all part of the COCO-80 label set, so pretrained detectors work out of the box.
| Image | Description, Source |
|---|---|
 (1376 x 768) |
A city-sidewalk scene with several pedestrians, parked cars, and a bus in the background — ideal for person / car / bus detection demos and NMS visualisations. Source: synthetic, generated with Google's Gemini 3 Pro Image (a.k.a. Nano Banana). |
 (1264 x 848) |
A wooden bowl on a wooden table containing whole apples, bananas, and oranges — for vision-language detection (e.g. "detect: apple, banana, orange") and open-vocabulary queries like "the leftmost fruit". Source: synthetic, generated with Google's Gemini 3 Pro Image. |
The SFM folder contains images that can be used for Structure From Motion (SFM).
In the subfolders you can find images of different scenes that can be used for SFM tasks.
Ellipse⬇️ (direct download of images as zip): scene showing the FH's ellipse. Images show a drone flight with 3 different heights and angles. Images are extracted from a video and subsampled. Note that the matching of the images is not perfect for all images.
An exemplary set of images is shown in the following table.
| Image 00 | Image 33 | Image 99 |
|---|---|---|
 |
 |
 |
The file tiny_nerf_data.npz is the dataset used by the 11b_TinyNeRF.ipynb tutorial — a small, self-contained Neural Radiance Field (NeRF) demo. It contains 106 views (100 × 100 RGB images), the corresponding 4 × 4 camera poses, and the focal length (≈138.89), all in a single NumPy .npz archive.
Originally sourced from Ben Mildenhall & Matthew Tancik's tiny_nerf.ipynb (cseweb.ucsd.edu/~viscomp/projects/LF/papers/ECCV20/nerf/tiny_nerf_data.npz); mirrored here for reliability.
⬇️ Direct link: tiny_nerf_data.npz
import numpy as np
data = np.load("tiny_nerf_data.npz")
images, poses, focal = data["images"], data["poses"], data["focal"]
print(images.shape, poses.shape, float(focal))
# (106, 100, 100, 3) (106, 4, 4) 138.8888788992210The fhhgb-hockey-dataset.zip contains a hockey-related dataset recorded at / for FH Upper Austria Hagenberg, used in computer vision exercises (e.g. detection / tracking on hockey footage). See the dataset author for the exact contents, splits, and licence.
⬇️ Direct link: fhhgb-hockey-dataset.zip
The pedestrian-detection-dataset.zip contains images of pedestrians crossing streets, annotated in YOLOv11 format for object detection. It features 752 images across train (592), valid (82), and test (78) splits, with 2 classes: crossing and waiting pedestrian. Annotations and splits follow the standard YOLO layout (<split>/images + <split>/labels, plus a data.yaml). A pretrained YOLO11n model (models/weights.pt) trained on this dataset is bundled inside the archive. Source: Roboflow Universe, licence CC BY 4.0.
⬇️ Direct link: pedestrian-detection-dataset.zip
In a Jupyter notebook (on Windows and Linux) you can use the following lines of code to download and unzip the entire zip-file of a dataset:
!curl -LJO "https://raw.githubusercontent.com/Digital-Media/cv_data/main/<dataset>.zip" --silent
import zipfile
with zipfile.ZipFile("<dataset>.zip", 'r') as zip_ref:
zip_ref.extractall(".")Replace <dataset> with hybrid_images, binary_leaves, thermal-rgb_ellipse, thermal-rgb_FH3, thermal-rgb_forest, thermal-rgb_hut, fhhgb-hockey-dataset, or pedestrian-detection-dataset.
For loose files (no zip needed) just curl or wget them directly, e.g.:
!curl -LJO "https://raw.githubusercontent.com/Digital-Media/cv_data/main/tiny_nerf_data.npz" --silent
!curl -LJO "https://raw.githubusercontent.com/Digital-Media/cv_data/main/object_detection/street.jpg" --silent