Skip to content
This repository has been archived by the owner on Oct 31, 2023. It is now read-only.

I have a question about json data in public. #3

Closed
rungjoo opened this issue Jun 18, 2021 · 4 comments
Closed

I have a question about json data in public. #3

rungjoo opened this issue Jun 18, 2021 · 4 comments

Comments

@rungjoo
Copy link

rungjoo commented Jun 18, 2021

I have a question about json data in public.

  1. What do each of the four values in bbox represent?
    • How do I change the bbox+position to (left, top, right, bottom)?
  2. Does relationships represent a relationship between indices?
  3. What is the difference between unique_id and index?

Thank you.

shanemoon added a commit that referenced this issue Jun 21, 2021
Information about Scene JSONs and bounding box values, #3
@satwikkottur
Copy link
Contributor

Hello @rungjoo ,

Thanks for your interest. We've added more clarity around these questions in data/README.md and included models/utils/visualize_bbox.py script to visualize bounding boxes. This should address your concerns, feel free to re-open this issue if you still have questions.

Thanks!

@tungngthanh
Copy link

I still confuse about the data. My aim is to know which object/scene corresponds to each dialogue.
So I take one example to verify whether I understand right:

{
'dialogue_idx': 11564, 
'domain': 'fashion', 
'mentioned_object_ids': [36, 42, 14, 21, 23, 26, 29],
 'scene_ids': {'0': 'm_cloth_store_1416238_woman_3_8', '6': 'm_cloth_store_1416238_woman_3_0'}
}

This is from the first data point in devtest dataset.
Is it right to interpret like that: from turn_idx 0 to turn_idx 5, we refer to the meta file "cloth_store_1416238_woman_3_8_scene.json" and the scene "cloth_store_1416238_woman_3_8.png"?

But I cannot find the first file in the data.
Did I miss something?

@satwikkottur
Copy link
Contributor

Hello @tungngthanh ,

Thanks for your interest.

Your interpretation about the scene files is correct.
turn_idx 0 to 5 use m_cloth_store_1416238_woman_3_8, while 6 onwards use m_cloth_store_1416238_woman_3_0.

Since these are multiple scene files per dialog, we have the m_ prefix. You should be able to find m_cloth_store_1416238_woman_3_8_scene.json file in the released JSON files (checked it at my end).

Please feel free to reopen the issue for follow-up questions.

@tungngthanh
Copy link

Hello @tungngthanh ,

Thanks for your interest.

Your interpretation about the scene files is correct.
turn_idx 0 to 5 use m_cloth_store_1416238_woman_3_8, while 6 onwards use m_cloth_store_1416238_woman_3_0.

Since these are multiple scene files per dialog, we have the m_ prefix. You should be able to find m_cloth_store_1416238_woman_3_8_scene.json file in the released JSON files (checked it at my end).

Please feel free to reopen the issue for follow-up questions.

Thank you for your reply. Follow your suggestions, I find that each scene idx corresponds to one json file and one image file.
However, now I face some problems when I mapped the object_local_id to the canonical object_id.
For example, the third dialogue (index 2) in the training set which has the scene_ids as follows:
{'0': 'm_cloth_store_1498649_woman_5_3',
'5': 'm_cloth_store_1498649_woman_5_9'}
its 7th uterance is :
{
'turn_idx': 6,
'system_transcript': "Sure, I'll add that now.",
'system_transcript_annotated': {'act': 'CONFIRM:ADD_TO_CART',
'act_attributes': {'slot_values': {}, 'request_slots': [], 'objects': [55]}},
'transcript': 'Actually, just add that brown jacket to my cart.',
'transcript_annotated': {'act': 'REQUEST:ADD_TO_CART',
'act_attributes': {'slot_values': {}, 'request_slots': [], 'objects': [55]}}
}
According to the document, we should expect to see the local_id 55 object in m_cloth_store_1498649_woman_5_9_scene.json, right?
However, when I load the file, I do not see 55 in index or unique_id.
Can you clarify it for me?

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants