Create environment with all requried packages
conda create -n --file req.txt
To go to the JSON format from datasets the following scripts can be used.
python ego_to_json.py
python cocohand_to_json.py # this script will merge coco hand and coco to get segmentations
From the JSON to the input formats and groundtruth
python json_to_extremenet.py # also for trident goes to COCO format
python json_to_groundtruth.py
python json_to_yolo.py
python mittel_to_groundtruth.py
The three different algorithms produce results in different formats:
Yolo unrelative maken
Standaard classname conf xywh
- Yolov5 - classname (1) , conf, xywh (relatief)
- Extremenet - classname (1), conf, xywh
- TridentNet - classname (1), conf, xywh
The results can be converted using Object-Detection-Metrics with the following command:
python create_results.py --gt ../../data/groundtruth/val --det ../../../experiments/experiment_1/validation_results/yolov5/ -detcoords rel -imgsize 1280,720 --start 0 --stop 131 --step 1
python create_results.py --gt ../../data/groundtruth/val --det ../../../experiments/experiment_1/validation_results/ExtremeNet/ --start 0 --stop 4550 --step 50 -np
python create_results.py --gt ../../data/groundtruth/val --det ../../../experiments/experiment_1/validation_results/trident/ --start 9 --stop 2549 --step 10 -np
For the testing:
python create_test_results.py --gt ../../data/groundtruth/test/ --det ../../../experiments/experiment_1/test_results_ego/extremenet/ -network extremenet -testset egohands
python create_test_results.py --gt ../../processed_data/mittel_zisserman/annotations/ --det ../../../experiments/experiment_1/test_results_mittel/trident/ -network tridentnet -testset mittal
~/networks/detectron2/tridentnet_training_output/*
~/networks/extremenet_sander/coco_extreme_train.pkl
~/networks/extremenet_sander/coco_extreme_val.pkl
~/networks/extremenet_sander/cache/nnet/ExtremeNet/*
~/networks/yolov5/runs/train/*
~/networks/yolov5/runs/detect/*