Hi @hunglc007,
I was going through the Dataset class source code (core.dataset.py), I stumbled upon this :
def load_annotations(self, dataset_type):
with open(self.annot_path, 'r') as f:
txt = f.readlines()
annotations = [line.strip() for line in txt if len(line.strip().split()[1:]) != 0]
np.random.shuffle(annotations)
return annotations
This line more specifically :
annotations = [line.strip() for line in txt if len(line.strip().split()[1:]) != 0]
This line of code is filtering negative samples (Images without the existence of the object). I may be missing something, somewhere. But, shouldn't negative samples be included in training ?
Hi @hunglc007, I was going through the Dataset class source code (
core.dataset.py
), I stumbled upon this :This line more specifically :
This line of code is filtering negative samples (Images without the existence of the object). I may be missing something, somewhere. But, shouldn't negative samples be included in training ?
Thank you in advance.