Custom Dataset

[johnolap]

Hello,

Thanks for this great project! I have a quick question; I'm adapting the CDN code to a custom dataset. However, I can't seem to understand the format. I have converted my dataset to the VCOCO format and I was able to extract the dict keys: (['image_id', 'ann_id', 'role_object_id', 'label', 'action_name', 'role_name', 'include']) but I don't understand the pattern in the trainval.json, train.json, val.json, and test.json.

[YueLiao]

Thx for your interest in our project~

We directly apply the JSON file provided by QPIC, and you may follow QPIC to transform the annotations with their script.

[johnolap]

Great. Thanks so much for your response.

[johnolap]

Hi, thanks again for your reply. I have a follow-up question, The convert_vcoco_annotations.py script converts vcoco to HOIA format. But I am trying to convert a coco format annotation to HOIA by manually adding the interactions. However, I'm unable to understand a few things, in HOIA format : what does {subject_id , category_id, object_id} mean in "hoi_annotations" dict key, what does {category_id} mean in "annotations" dict key"

[YueLiao]

Hi, the subject_id and object_id is the index of the box annotations list. And category_id in annotations means the category of this box.

[johnolap]

Got it, thank you!