# Image Classification for Fiber Box Installation — Data

来源：https://www.kaggle.com/competitions/ada-image-recognition-fiber/data

采集时间：2026-10-19T06:34:59Z

The dataset contains 1435 images split in train and test.
The annotations are provided in the YOLO format. All pictures having annotations have a corresponding *.txt* file containing the coordinates of these annotations. Note that pictures with no annotation don't have a corresponding *.txt* file. Note that no *.yaml* file is provided so that you can make your own train/val split.

```
<dataset>/
    images/
        train/
            <uuid1>.<ext>
            <uuid2>.<ext>
            ...
        test/
            <uuid3>.<ext>
            <uuid4>.<ext>
            ...
    labels/
        train/
            <uuid1>.txt
            <uuid2>.txt
            ...
```
The *.txt* files contain:
- one row per object: 
- format: class (starting from 0), x_center, y_center, width, height
- Box coordinates are in normalized xywh format (from 0 - 1).

Here is the number of available annotations:

![](assets/b5410362a7dfea68b380a82d.png)

The annotations are given under the form of segmentation instances, as in the following example:

![](assets/c04c194d8fc243d332ad4b4e.png)

There are different types of views in the dataset. Note that it's not always possible to see all the components in the image. In some cases, no component is visible on the image, as illustrated below: 

![](assets/6de991585eaef6b556b89e20.png)

Finally, there might be some inconsistencies in the dataset, such as the missing left rubber and tie-wrap in the following example. An iterative approach to dealing with labeling mistakes can maybe be considered.

![](assets/5aa6840760afff48183f4888.png)