xf.jian

I also want to train hands-only, but the dataset provided is hands + arm. Any news on the topic ? @owenzlz

Refer to the console logs, maybe the config file is not loaded or set correctly. Check `config.xml`, the camera index should be pre-set there. But I'm sorry I haven't focus...

Glad to hear that. The core algorithm of guidance is modified above the [monocular guidance system](https://github.com/pengsongyou/CalibrationWizard), you can check the paper for basic principles.

As I known, metric3d generate pointcloud with single image and depth map, while DROID-SLAM do it with all images and trajectory. If the pointcloud by droid+metric looks bad, I think...

Hi, as not enough infomation is provided to solve the question, check that first: #9 BTW, too much moving objects will also makes it worse.

Metric3D is an amazing work on metric depth estimation, which helps me finish my graduation project. THANKS for your great work and THANKS for linking!

Hi， It is possible to estimate the volume of objects as the volume size is real, just as that in Metric3D. But you'll need to design an algorithm to perform...

Hi, It looks like that the console output has no errors. For the mesh output, the main architecture was successfully reconstructed but lacks details. There might be some problems: -...

Hi, It seems that the problem comes from metric3d. I'm not sure if it works well with `metric_depth_vit_giant2_800k.pth` on your device. If that weights work, you can check the code...

Hi， Before running `run.py`, you should run `depth.py` for depth estimation first. I don't make it simultaneous with SLAM thread. For the warning, I'm sorry I forgot update the code...