|
--- |
|
task_categories: |
|
- image-classification |
|
language: |
|
- en |
|
tags: |
|
- FGVC |
|
pretty_name: BOATS |
|
size_categories: |
|
- 10K<n<100K |
|
--- |
|
# Dataset Card for BOATS Dataset |
|
|
|
## Dataset Description |
|
|
|
- **Homepage:** |
|
- **Repository:** |
|
- **Paper:** |
|
- **Leaderboard:** |
|
- **Point of Contact:** |
|
|
|
### Gaining Access |
|
If you want to gain access to the dataset send me a message on LinkedIn www.linkedin.com/in/cringgaard |
|
|
|
### Dataset Summary |
|
|
|
A dataset consisting of over a 100.000 images of 7108 classes of sailboats from small dinghies to large ships. The hull type, rigging, construction material and ballast type have also been noted. |
|
|
|
This dataset was used for my bachelor's thesis in data science and machine learning at the university of Copenhagen. |
|
The images were scraped from the web as specified in the [thesis](Fine_Grained_Visual_Categorisation_of_Boats.pdf). |
|
|
|
### Abstract |
|
Modern computer vision models are trained on large datasets but typically require fine-tuning to perform |
|
well on specific tasks. These tasks are often valuable on their own but can also be used for evaluating the |
|
pre-trained representations. In this thesis one such dataset is presented, a fine-grained sailboat dataset |
|
with annotated attributes. The dataset is suitable as starting point for the automation of classifying |
|
sailboats, which can be of value for e.g., monitoring marinas and sensitive waters. The data also presents |
|
new challenges to computer vision models in how it requires the classification of attributes that are not |
|
directly visible due to being submerged in water. While there is a limited amount of images available for |
|
each boat type, collectively the different boat attributes are present in multiple images. To further expand |
|
the dataset, a weakly labeled extension is scraped via an image search on a popular search engine. We |
|
train a variety of classifiers over the dataset and explore if the models benefit from learning to classify |
|
specific attributes in combination with others. Our findings show that using such multitask learning and |
|
a weakly labelled large dataset, it is possible to create a model capable of inferring the exact boat type at |
|
least as well as a human with expertise in the area. |
|
|
|
|
