search

GAP-LAB-CUHK-SZ / Total3DUnderstanding

Implementation of CVPR'20 Oral: Total3DUnderstanding: Joint Layout, Object Pose and Mesh Reconstruction for Indoor Scenes from a Single Image
MIT License
419 stars 50 forks source link

Camera intrinsic parameters and distortion coefficients #52

Closed qijiahui7 closed 9 months ago

qijiahui7 commented 1 year ago

I want to run my own captured photos using a demo. Following the steps, I input the bounding box coordinates (xmin, ymin, xmax, ymax) and label, as well as the camera intrinsic parameters and the image. However, I found that the 3D reconstruction result is not satisfactory, and the bounding boxes for objects like tables and chairs are not accurately selected. I'm not sure if it's because I didn't input the camera distortion or some other unknown reasons. I hope to get a response. Thank you.

yinyunie commented 1 year ago

Hi,

Since our model is trained on SUNRGBD, I am not sure if your camera intrinsics and image size are similar to the cameras in this dataset. If not, there could be a domain gap and I would recommend you to rescale your image and camera intrinsics to the same as in the dataset.