This repository hosts utility scripts for image warping research.
Key files are summarized in this readme
Consult the python files in each subfolder for further details.
Obtain coco-format json of specific weather or tod
(1) Convert bdd-format to coco-format: bdd/bdd2coco.py
(2) Filter coco-format json based on weather and tod: bdd/filter_file.py
Generate json containing bboxes for instance-level warping:
(1) Run: sem_seg/seg_to_bbox.py
Visualize bboxes (and get results) on images
(1) Run detection on image
(2) Get category-wise results: coco/coco_mAP_simple.py
(3) Get bboxes visualizations: coco/vis_det_each.sh
Visualize bboxes on videos
(1) Convert video to image: video/video2image.py
(2) Generate pseduo-json for images: jsons/create_empty_json.py
(3) Run detection on image
(4) Visualize detected images: coco/vis_det_each.sh
(5) Merge images to video: video/image2video.py
Obtain 2d coco annotations from the DENSE dataset
(1) Get coco-format jsons with 2D bboxes: dense_3d2d/gen_coco.py
(2) Visualize 2D bboxes for debug: dense_3d2d/vis_many.py
(3) Count the occurence for each category: dense_3d2d/count_category.py
(4) Split into train/test based on ratio and tod constraints: dense_3d2d/train_test_split.py