Open WangYixuan12 opened 1 year ago
Hi Yixuan, the masks we provided may not be very accurate as we used an off-the-shelf instance segmentation model. They are offered for convenience only, and you are not required to use them. We suggest using a better segmentation model to obtain more precise masks, which can lead to better results in pose estimation.
Hi Yang! Thank you for creating such a great dataset. I am using this dataset in my project. However, I found that some mask images are missing in the dataset (e.g. camera/0003/2021-08-31--14-39-07/images/{288~294}-mask.png). I wonder whether they are missing by mistake or for a particular reason. I also notice that some mask images do not make sense (e.g. camera/0003/2021-08-31--14-39-07/images). Some mask images are just simple masks or a small portion of objects. Is that by mistake or for a particular reason?