• add new coco utils. check COCO.md for example usages.
• add slicing notebook.
• fix some mmdet/numpy warnings.

• add new coco utils (https://github.com/obss/sahi/pull/42). check here for example usages.

• refactorized Coco class, added CocoVid class

• added coco_error_analysis.py script:

python scripts/coco_error_analysis.py results.json output/folder/directory --ann coco/annotation/path

will calculate coco error plots and export them to given output folder directory.

If you want to specify mAP result type, set it as --types bbox mask.

If you want to export extra mAP bar plots and annotation area stats add --extraplots argument.

If you want to specify area regions, set it as --areas 1024 9216 10000000000.

• fix a sliced prediction bug happening on small images

Major Updates

• refactorize main predict function:

python scripts/predict.py --visual --source image/file/or/folder --model_path path/to/model --config_path path/to/config

will export predictions visuals to runs/predict/exp folder.
If you want to export prediction pickles and cropped predictions add --pickle and --crop arguments.

• add option to export coco json:

python scripts/predict.py --coco_file path/to/coco/file --source coco/images/directory --model_path path/to/model --config_path path/to/config

will perform inference using provided coco file, then export results as a coco json file to runs/predict/exp/results.json

• add script for coco2yolov5 conversion:

python scripts/coco2yolov5.py --coco_file path/to/coco/file --source coco/images/directory --train_split 0.9

will convert given coco dataset to yolov5 format and export to runs/coco2yolov5/exp folder .

Minor Updates

• add export_format to crop_object_predictions
• fix coco annotation serialization
• more generic coco handling
• add option to export visuals
• update ci/package testing workflows

• fix cli argument

• fix function argument

• add export format selection

• add coco to yolo conversion utility

• added merger for single predictions

• match_iou_thr is configurable
• better time profiling