Can't get a right nerf resualt

[nypyp]

Hi, I'm confusied when I run orb_slam3_ros with nerf_brige, the viewer 's image stay still in the center and never have another image shows up, and the nerf nerver work good, is there anything I do wrong? Or is there any rosbag could verify the configuration is right?

[Eric-Ho-Matrix]

Looks like we are facing the same problem. My implementation also have this prob, Nerfstudio basically stuck at the first several images and I also cannot get a fairly good model.

[javieryu]

Looks like your images are grey scale? Maybe this has something to do with it?

Can you give some more details on your setup?

Also, due to issues with the Nerfstudio Viewer there is no easy way to update the image locations in the viewer as they come in. So while images and poses are being added to the training set they may not appear in the viewer.

[bharath-k1000]

Hi, I am also facing the same issue. The input image stream is not graycaled for me but i am getting the same result in viewer. the topic /camera/color/image_raw has rgb image data and is written correctly in the .json file.

I am using an an Intel realsense d435 rgbd camera (no imu).

Orbslam3 seems to be working fine seeing the the features that are tracked and the pointcloud. But nerf generation seems to be stuck.

• ros 1
• ubuntu 20.04.

[bharath-k1000]

UPDATE:
Issue was resolved by changing the camera parameters in the .json file.

[lvmingzhe]

I'm sorry for replying in a closed issue, because I encountered the same problem. The dataset comes from tum_rgbd_dataset_freiburg1_xyz, and I used thien94 orb_slam3_ros for pose estimation, configuring the corresponding ns_orb3_tum.json

{
	"fx": 525.0,
    	"fy": 525.0,
    	"cx": 319.5,
    	"cy": 239.5,
    	"k1": 0.0,
    	"k2": 0.0,
    	"k3": 0.0,
    	"p1": 0.0,
    	"p2": 0.0,
    	"H": 480,
    	"W": 640,
	"image_topic": "/camera/rgb/image_color",
	"pose_topic": "/orb_slam3/camera_pose"
}

After then , I continue to run by step 1 .

roslaunch orb_slam3_ros tum_rgbd.launch

step 2 .

rosbag play ~/data/rgbd_dataset_freiburg1_xyz.bag

step 3.

python ros_train.py --method_name ros_nerfacto --data /home/hello/code/nerf_bridge/ns_orb3_tum.json --pipeline.datamanager.data_update_freq 1.0

finally got this result looks not good.

could you please give me some suggestions? Thanks.

[lvmingzhe]

I solve this problem by following thien94/orb_slam3_ros#7 (comment)
change RGB.DepthMapFactor to 1.0 in TUM<version>.yaml

[Augusthyq]

UPDATE: Issue was resolved by changing the camera parameters in the .json file.

Sorry, I have encountered the same problem. How you change your json file?

[javieryu]

@Augusthyq Happy to help if you can open up a new issue, and include your setup details.