search

shikras / d-cube

A detection/segmentation dataset with labels characterized by intricate and flexible expressions. "Described Object Detection: Liberating Object Detection with Flexible Expressions" (NeurIPS 2023).
https://arxiv.org/abs/2307.12813
Other
104 stars 7 forks source link

Some images have no associated annotations? #5

Closed ImcwjHere closed 9 months ago

ImcwjHere commented 10 months ago 
images = pickle.load(open("images.pkl", "rb"))
count = 0
for image in images.items():
    anno_id = image[1]["anno_id"]
    if not len(anno_id):
        count += 1     
# Output: count = 2690

Are there 2690 images that don't have a corresponding anno_id?

Charles-Xie commented 9 months ago

Hi, Thanks for your interest in our work! We appreciate that very much. About your question, the answer is yes. There are some images that does not have a corresponding anno_id. These images contain no positive instances of the 422 categories annotated in our dataset, i.e., they are negative samples for all the categories.

Charles-Xie commented 9 months ago

I think this issue is resolved so I will close it now. Please feel free to reopen it if the question is not clarified or open another issue if you have any other question. We are very thankful for your attention to our work.

hhaAndroid commented 9 months ago

@Charles-Xie What are negative samples used for? Are they useful during evaluation?

Charles-Xie commented 8 months ago

@Charles-Xie What are negative samples used for? Are they useful during evaluation?

Sorry for the late reply. Negative samples are also used during evaluation. For a category, the ideal model should reject any predictions on the negative samples. This verifies an important aspect of detection models, i.e., rejecting negative samples and avoiding FP predictions.