How to generate center and bounding box for each depth image?

[tiaopao]

I'm appreciating your great work.

For my own human pose dataset, in order to train the my own model:

1. How to generate the center and bounding box for each depth image?
2. How to generate the mean/standard deviation for each depth image?

For above questions, would you please share the method or algorithm?

[zhangboshen]

Hi,

1. there are few choices that you can try, Faster RCNN (or whatever detector) human detector for bbox generation, and centers are borrowed from Gyeongsik et al. (https://github.com/mks0601/V2V-PoseNet_RELEASE) in our setting, but you can definetely train a simple center detector yourself.
2. refer to this issue: Code for calculating mean and std #29 for the discussion of mean/std computaion, you can easily compute your own mean/std w.r.t your dataset.

[hankhuynh1011]

Hi Boshen, I appreciate your work. I would like to reproduce it on the iTop dataset, but I have a problem with the bounding box.
Most of the Detector (faster rcnn, ..) work on RGB images, so we need to finetune the Detector with depth dataset, right? and what is the depth dataset you used to finetune the Detector model?

[Shreyas-NR]

Hi @hankhuynh1011 ,

1. Were you able to utilize this model to predict the Joints for a custom dataset?
2. I'm also trying to pass a depth frame along with the ITOP side dataset and change the mean value so that the input depth frame to the model matches with the ITOP_side dataset. Unfortunately, the results are very bad.
3. Could you tell me if you were able to do something more on this?

[zhangboshen]

eed to finetune the Detector with depth dataset, right? and what is the depth dataset you used to finetune the Detector model?

We finetune a FRCNN detector on the ITOP training set, GT bndbox is generated by calculate the min/max values of 2D GT keypoints.

[ProVival]

Hi @hankhuynh1011 ,

1. Were you able to utilize this model to predict the Joints for a custom dataset?
2. I'm also trying to pass a depth frame along with the ITOP side dataset and change the mean value so that the input depth frame to the model matches with the ITOP_side dataset. Unfortunately, the results are very bad.
3. Could you tell me if you were able to do something more on this?

Were you able to solve these problems :D i am having same issues