Repository navigation
[Docs] Code to process data into a format usable by the demo #14
Description
Branch
main branch https://mmdetection3d.readthedocs.io/en/latest/
📚 The doc issue
Hi, can you please post the code that processes the data into a usable format for the demo?
Suggest a potential alternative/fix
No response
Activity
LZ-CH commentedon Mar 20, 2024
Such as camera.json.
mxh1999 commentedon Mar 20, 2024
I apologize for omitting the raw data of our open-world demo from the distribution, we'll add it as soon as possible. Please refer to demo.ipynb for instructions on how to use it.
Thanks again for pointing this out!
LZ-CH commentedon Mar 20, 2024
Okay, thank you very much. I also want to ask, in the 3D detection task, before the data is input to the model, it will be scaled. During the model inference stage, the coordinates of prediction will be scaled and affine transformed to the original coordinate system? Because I didn’t find the relevant code for this step, so I want to ask this question!
Thank you again for this great work.
mxh1999 commentedon Mar 21, 2024
I'm a bit unclear about the specific steps you're referring to. If you're asking about the image processing steps before the backbone, you can refer to the Det3DDataPreprocessor module. It handles the necessary pre-processing of the images.
On the other hand, if you're inquiring about the transformation of predicted boxes from ego-centric coordinates to global coordinates, you can look into the single_scene_multiclass_nms function within the bbox_head module. It handles the post-processing and transformation of the predicted bounding boxes.
LZ-CH commentedon Mar 21, 2024
Thank you very much. With your help, I found the data post-processing process. Because I want to implement a 3D detector for use in real scenes based on your work, I pay more attention to the data input and output format of the demo. Looking forward to your additions on demo data preparation.
Thank you again for your help！
mxh1999 commentedon Mar 22, 2024
@LZ-CH We have uploaded the raw data of our open-world demo, download it from Google Drive or BaiduYun.
LZ-CH commentedon Mar 22, 2024
@mxh1999 Thank you for your assistance. I'm thinking about trying to scan a room like you do in OpenScan using an RGB-D camera (Depth Camera D435i). Could you possibly recommend any projects or academic papers that might guide me in generating my own environment ?
mxh1999 commentedon Mar 22, 2024
We use Kinect to generate the mesh of the environment. If you want to work with your own data, this may be helpful.
LZ-CH commentedon Mar 22, 2024
@mxh1999 Thank you for solving my problem and wish you a happy life!