---
dataset_info:
  features:
  - name: image_id
    dtype: int64
  - name: image
    dtype: image
  - name: objects
    struct:
    - name: bbox
      sequence:
        sequence: int64
    - name: categories
      sequence: string
  splits:
  - name: train
    num_bytes: 70990022.0
    num_examples: 566
  download_size: 70811624
  dataset_size: 70990022.0
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---
---
# Dataset Card for "oct-object-detection-v4-merge"
Dataset is composed of images with multiples object detection box in coco format (xmin, ymin, xmax, ymax). Images are OCT (type of eye scaner) with boxes indicating some features associated to AMD disease. 
The difference from v3 is images are grouped (not duplicated images in multiples row) and they can have multiples labels-boxes in the objects field. So there are, 566 unique images, there are 566 rows, one per image.
Also, overlapped boxes are joined as merge function
[Source datataset](https://doi.org/10.1101/2023.03.29.534704)