Using the LightGlue feature extractor, implemented a Visual Odometry system to estimate the camera's motion in a 3D environment.
cd simple_vo_challenge
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
python read_coco.py
python visual_odometry.py
python animate_plot.py
xdg-open animation.gif
The provided dataset (sample_coco_dataset.json
) is formatted in the COCO style. It contains:
images
: Each image entry is associated with anid
,width
,height
, andfile_name
.annotations
: Annotations link detected objects with their bounding boxes in specific images. Each annotation entry has animage_id
,bbox
(which provides the bounding box in the format[x, y, width, height]
), and acategory_id
.