cfg/cfg.yml

error_analysis:
  # detection_classes: ["person", "bicycle"]
  labels_dict: {"person": 1, "bicycle": 2}
  conf_thresholds: [0.2, 0.35, 0.5, 0.65, 0.8]  # some call it score threshold
  iou_thresholds: [0.2, 0.35, 0.5, 0.65, 0.8]   # back in my day we call it NMS threshold *shakes fist*
  # nms_thresholds: [0.2, 0.5, 0.8] 
  bbox_format: "pascal_voc"                     # yolo / coco / pascal_voc (WIP feature)
  peekingduck: True                             # False if using your own model for inference without peekingduck wrapper, else True
  ground_truth_format: "coco"                   # yolo / coco / pascal_voc (WIP feature)
  idx_base : 1                                  # to indicate whether the class index is zero or one based. Applies to both GT and pred class

pkd:
  model: "yolo"
  yolo_ver: "v4tiny"

dataset:
  classes: ["person", "bicycle"]                # same as ['general']['detection_classes'] field above
  img_folder_path: 'data/annotations_trainval2017/coco_small/' # relative path from root for saving the coco dataset images
  annotations_folder_path: 'data/annotations_trainval2017/annotations/' # relative path from root to the annotations file
  annotations_fname: "instances_val2017.json"    # what is the name of your json file?

visual_tool:
  bbox_thickness: 2                             # how thicc you want the bbox to be
  font_scale: 1                                 # how big you want the fonts to be 
  font_thickness: 2                             # how thicc you want the fonts to be  
  pred_colour: [255, 0, 0]                      # prediction colour, [B,G,R]
  gt_colour: [0, 255, 0]                        # Ground truth colour, [B,G,R]