zubair-irshad / CenterSnap

Pytorch code for ICRA'22 paper: "Single-Shot Multi-Object 3D Shape Reconstruction and Categorical 6D Pose and Size Estimation"
https://zubair-irshad.github.io/projects/CenterSnap.html
286 stars 47 forks source link

Questions about the making of custom datasets #31

Open junho-choi-div opened 9 months ago

junho-choi-div commented 9 months ago

Hello, I am currently in the process of training the Centersnap model by establishing a custom data set with the intel realsense D435i camera. I thought I formed the NOCS MAP and other necessary datasets well, but the prediction of 6d pose estimation keeps going wrong. In order to check if the data set construction pipeline I made works well, I created a Nocs map for the bottle class of the Train data set and built the data set and conducted the training, and I got the desired result as follows. Leave a question about what parts other than the internal parameters of the camera need to be changed. Thanks

0_image 0_peaks_output box3d0 projection0

zubair-irshad commented 8 months ago

Thanks for your interest. It looks like you are trying to create a custom dataset for CenterSnap fine-tuning but you shared an image from NOCS Real dataset. Are these results with the finetuning or zero-shot evaluation?

Can you share examples of how you created your custom dataset?