Fall 2024 Computer Vision Final Project Data

This repository contains the cleaned up data for the fall 2024 Computer Vision final project. To access the data, clone this repository on the machine where you train the model. This repository is fairly large, so it might take a few minutes to clone.

Data Organization

All data are stored in the data/ folder, the organization is as follows:

data/
├── Pair_S_000001_3503_3563/
│   ├── S_000001_frame_3503.png
│   ├── S_000001_frame_3503.annotation.txt
│   ├── S_000001_frame_3563.png
│   ├── S_000001_frame_3563.annotation.txt
├── Pair_S_000001_3563_3623/
│   ├── S_000001_frame_3563.png
│   ├── S_000001_frame_3563.annotation.txt
│   ├── S_000001_frame_3563.png
│   ├── S_000001_frame_3623.annotation.txt

It is guaranteed that each subdirectory in the data folder contains at least 4 files: a pair of PNG image files and a pair of annotation txt files. Each subdirectory may contain more than 4 files (such as the bounding box images), but we don't have to worry about those.

It is also guaranteed that the images and annotations are named as *_aaaa.png, *_aaaa.annotation.txt, *_bbbb.png, and *_bbbb.annotation.txt. The * represent 0 or more of any character; aaaa and bbbb represent two different 4-digit frame numbers.

It is guaranteed that the labels are in the correct format (contains 8 space-separated numbers).

It has NOT be verified whether the PNGs have the same dimension, so a transform pipeline before training is necessary. It is also NOT verified whether the annotations are correct. The hope is that a few incorrect annotations won't affect the training result too much.

Data Index

A file named index.txt in provided in the root directory of this repository. Each row of the file contains 4 comma-separated values: 4 relative paths to a pair of images and their annotations. In the following order:

[path to frame 1], [path to annotation 1], [path to frame 2], [path to annotation 2]

Here is an example of what the file looks like:

data/Pair_S_000001_3503_3563/S_000001_frame_3503.png,data/Pair_S_000001_3503_3563/S_000001_frame_3503.annotation.txt,data/Pair_S_000001_3503_3563/S_000001_frame_3563.png,data/Pair_S_000001_3503_3563/S_000001_frame_3563.annotation.txt
data/Pair_S_000001_3563_3623/S_000001_frame_3563.png,data/Pair_S_000001_3563_3623/S_000001_frame_3563.annotation.txt,data/Pair_S_000001_3563_3623/S_000001_frame_3623.png,data/Pair_S_000001_3563_3623/S_000001_frame_3623.annotation.txt

Iterating through this file line-by-line should simplify the data loading process.