README.md

VIPriors Object Detection challenge - Data

The VIPriors object detection challenge uses DelftBikes dataset. In particular, we use the bounding box annotations.

Please follow the instructions below to set up the data for this challenge. To submit your contribution to our challenge generate your models predictions over the test set and submit the predictions according to the instructions in the main README.

We also provide a validation set derived from training set. Validation results can be submitted to Development(Validation set). For final submission, you can use both training and validation sets for training.

Setting up data

These are the instructions for setting up the data for the VIPriors object detection challenge.

1. Download the DelftBikes images from DelftBikes;
2. Extract the ZIP file to data. This creates folders data/train, data/test, train_annotations.json and fake_test_annotations.json. fake_test_annotations.json is used just for generating submission.

Now you are ready to use the data.

• The root directory of the training and testing images is data/DelftBikes;
• The annotations files to be used are:
  • Train set (8K images): data/annotations/train_annotations.json
  • Fake Testing set (2K images, no correct labels provided): data/annotations/fake_test_annotations.json
  • To generate validation and new train set (both are derived from original training set), you can run valset_generation.py which is located under data directory. The script moves 1000 train images to val directory and produces new annotations:
    • val directory consists of 1000 images.
    • train directory consists of 7000 images.
    • val_annotations.json and new_train_annotations.json.