longzw1997 / Open-GroundingDino

This is the third party implementation of the paper Grounding DINO: Marrying DINO with Grounded Pre-Training for Open-Set Object Detection.
MIT License
420 stars 65 forks source link

about train dataset questions #47

Closed xiyangyang99 closed 10 months ago

xiyangyang99 commented 10 months ago

There was an issue converting my local Coco dataset to a. jsoll file using the script you provided. My dataset is just one category. I have made two modifications as follows: QQ图片20231222102850 QQ图片20231222102857

This is the information output after running the script: QQ图片20231222102903

The output log of the training process prompts the following issues:

1703212340580

Looking forward to your reply,thinks!

BIGBALLON commented 10 months ago
  1. In dictionary form, indices start from "0" (it is essential to start from 0 to accommodate caption/grounding data). Here is an example
  2. check https://github.com/longzw1997/Open-GroundingDino/blob/main/tools/v3det2odvg.py, which is the script that converts the coco format dataset to ogvd format.
xiyangyang99 commented 10 months ago

Thank you for your reply. May I ask if we will proceed with the Tensorrt acceleration work for GroundingDINO

---Original--- From: "WILL @.> Date: Sat, Dec 23, 2023 15:22 PM To: @.>; Cc: @.**@.>; Subject: Re: [longzw1997/Open-GroundingDino] about train dataset questions(Issue #47)

In dictionary form, indices start from "0" (it is essential to start from 0 to accommodate caption/grounding data). Here is an example

check https://github.com/longzw1997/Open-GroundingDino/blob/main/tools/v3det2odvg.py, which is the script that converts the coco format dataset to ogvd format.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

BIGBALLON commented 10 months ago

Maybe you need to contact the original author https://github.com/IDEA-Research/GroundingDINO