from typing import Union
import cv2
from .. import functions as f
class Error:
""" A base class for all error types. """
def fix(self) -> Union[tuple, None]:
"""
Returns a fixed version of the AP data point for this error or
None if this error should be suppressed.
Return type is:
class:int, (score:float, is_positive:bool, info:dict)
"""
raise NotImplementedError
def unfix(self) -> Union[tuple, None]:
""" Returns the original version of this data point. """
if hasattr(self, 'pred'):
if self.pred['used'] is None: return None
else: return self.pred['class'], (self.pred['score'], False, self.pred['info'])
else:
return None
def get_id(self) -> int:
if hasattr(self, 'pred'):
return self.pred['_id']
elif hasattr(self, 'gt'):
return self.gt['_id']
else:
return -1
def show(self, dataset, out_path:str=None,
pred_color:tuple=(43, 12, 183), gt_color:tuple=(43, 183, 12),
font=cv2.FONT_HERSHEY_SIMPLEX):
pred = self.pred if hasattr(self, 'pred') else self.gt
img = dataset.get_img_with_anns(pred['image_id'])
if hasattr(self, 'gt'):
img = cv2.rectangle(img, *f.points(self.gt['bbox']), gt_color, 2)
img = cv2.putText(img, dataset.cat_name(self.gt['category_id']),
(100, 200), font, 1, gt_color, 2, cv2.LINE_AA, False)
if hasattr(self, 'pred'):
img = cv2.rectangle(img, *f.points(pred['bbox']), pred_color, 2)
img = cv2.putText(img, '%s (%.2f)' % (dataset.cat_name(pred['category_id']), pred['score']),
(100, 100), font, 1, pred_color, 2, cv2.LINE_AA, False)
if out_path is None:
cv2.imshow(self.short_name, img)
cv2.moveWindow(self.short_name, 100, 100)
cv2.waitKey()
cv2.destroyAllWindows()
else:
cv2.imwrite(out_path, img)
def get_info(self, dataset):
info = {}
info['type'] = self.short_name
if hasattr(self, 'gt'):
info['gt'] = self.gt
if hasattr(self, 'pred'):
info['pred'] = self.pred
img_id = (self.pred if hasattr(self, 'pred') else self.gt)['image_id']
info['all_gt'] = dataset.get(img_id)
info['img'] = dataset.get_img(img_id)
return info
class BestGTMatch:
"""
Some errors are fixed by changing false positives to true positives.
The issue with fixing these errors naively is that you might have
multiple errors attempting to fix the same GT. In that case, we need
to select which error actually gets fixed, and which others just get
suppressed (since we can only fix one error per GT).
To address this, this class finds the prediction with the hiighest
score and then uses that as the error to fix, while suppressing all
other errors caused by the same GT.
"""
def __init__(self, pred, gt):
self.pred = pred
self.gt = gt
if self.gt['used']:
self.suppress = True
else:
self.suppress = False
self.gt['usable'] = True
score = self.pred['score']
if not 'best_score' in self.gt:
self.gt['best_score'] = -1
if self.gt['best_score'] < score:
self.gt['best_score'] = score
self.gt['best_id'] = self.pred['_id']
def fix(self):
if self.suppress or self.gt['best_id'] != self.pred['_id']:
return None
else:
return (self.pred['score'], True, self.pred['info'])
