search

poodarchu / learn_aug_for_object_detection.numpy

Learning Data Augmentation Strategies for Object Detection in numpy
55 stars 8 forks source link

bug of _apply_multi_bbox_augmentation func #7

Open dypark86 opened 4 years ago

dypark86 commented 4 years ago

when i use _apply_multi_bbox_augmentation function with func_changes_bbox=false, this function shuffle the bbox order. but label order is not shuffle. I think, this can make problem of mislabeling.

dypark86 commented 4 years ago

np.random.shuffle(bboxes) change the order of bboxes. so below code should be changed with np.random.permutation()

if not func_changes_bbox: np.random.shuffle(bboxes) loop_bboxes = bboxes -> if not func_changes_bbox: loop_bboxes = np.random.permutation(bboxes)

Haijunlv commented 4 years ago

np.random.shuffle(bboxes) change the order of bboxes. so below code should be changed with np.random.permutation()

if not func_changes_bbox: np.random.shuffle(bboxes) loop_bboxes = bboxes -> if not func_changes_bbox: loop_bboxes = np.random.permutation(bboxes)

i think this is the key reason why this code can not reproduce tensorflow result and i will fix and try

Jacobew commented 4 years ago

@Haijunlv Hi, any update here?