Error when Testing on SUNRGB-D dataset
Closed this issue · 4 comments
Hi, I am trying to run "python inference_sunrgbd.py --config configs/train_sunrgbd.yaml", but got the following error:
It seems that the paths "/home/newdisk/merge-data" and "/home/newdisk/BigData/sunrgbd_det_results/visualization" in the train_sunrgbd.yaml file are missing. Could you please help with the problem?
Thank you very much! Looking forward to your kind reply.
Hi! Sorry, I forgot to change the SUNRGBD config and upload the detection results for SUNRGBD data, thank you very much for your reminder! I have already updated the code and SUNRGBD data, please update your local files.
Hi! Sorry, I forgot to change the SUNRGBD config and upload the detection results for SUNRGBD data, thank you very much for your reminder! I have already updated the code and SUNRGBD data, please update your local files.
Hi, thank you very much for your quick reply. I checked the code and files, it worked now!
btw I see you train and test on Pix3D and FRONT3D dataset in the paper. I am wondering whether you have trained the model on SUNRGB-D dataset. If so, could you please also provide the data-preprocessing procedure and the SUNRGB-D training code?
We didn't train on SUNRGBD, because SUNRGBD doesn't provide GT mesh, only point cloud is provided. And the point cloud is not accurate enough for individual objects so it's a little difficult for us to generate GT SDF.
A more important reason is that we want to show the generalization ability of our model on SUNRGBD, so we directly use the model trained on FRONT3D to test on SUNRGBD.
You can refer to Total3D to process the SUNRGBD dataset.
Okay! Sorry that I forgot SUNRGBD dataset doesn't provide GT mesh. Have a nice day!