Hi,

Thanks again for your great work. I have a question about the implementation of 2D consistency loss https://github.com/shangbuhuan13/SO-Pose/blob/a3a61d2c97b1084a4754d6c12e45e16d85809729/core/gdrn_selfocc_modeling/losses/crosstask_projection_loss.py#L158

I am confused why the loss is divide by 572.3. In datasets/BOP_DATASETS/lm/camera.json I see the camera information is

{
  "cx": 325.2611,
  "cy": 242.04899,
  "depth_scale": 1.0,
  "fx": 572.4114,
  "fy": 573.57043,
  "height": 480,
  "width": 640
}

Also, will this impact YCBV dataset, since it has different camera intrinsic parameters? Thanks!

Hi,

Thanks for your great work! I have a question about the backbone in LM datasets. In your paper, you said that "As backbone we leverage ResNet34 [6] for all experiments on the LM dataset", but the config files in this repo seems different:

       BACKBONE=dict(
            FREEZE=False,
            PRETRAINED="mmcls://resnet50_v1d",
            INIT_CFG=dict(
                _delete_=True,
                type="mm/ResNetV1d",
                depth=50,
                in_channels=3,
                out_indices=(3,),
            ),
        ),

and the output feature dimension is 2048, not 512 as in GDR-Net.

Could you help to clarify this? Thanks!

您好，请问作者可以再提供一次lib.egl_renderer吗，其他以前的网盘都过期了，十分感谢。 https://github.com/shangbuhuan13/SO-Pose/blob/b1cfa9c20bfbb0b4ccd1e8c421e392958628aa4f/core/gdrn_selfocc_modeling/tools/lmo/lmo_2_vis_poses.py#:~:text=from%20lib.egl_renderer.egl_renderer_v3%20import%20EGLRenderer

如题目所示，请问作者可否提供下可视化结果时用到的lib.egl_renderer，无比感谢。 https://github.com/shangbuhuan13/SO-Pose/blob/b1cfa9c20bfbb0b4ccd1e8c421e392958628aa4f/core/gdrn_selfocc_modeling/tools/lmo/lmo_2_vis_poses.py#:~:text=from%20lib.egl_renderer.egl_renderer_v3%20import%20EGLRenderer

您好! 我准备使用gdrn_selfocc_multistep_40E.py配置文件复现实验，但是我发现在生成xyz后lm/train_pbr/xyz_crop已经很大，并且准备生成train_pbr/Q0,但是这个我估计要3T的存储空间，所以想问下配置文件中TRAIN2=("lmo_pbr_train",),的这个是否必要?

如图所示，cpu无法跑满，导致速度很慢，请问这个有解决办法吗。不知道是不是因为你们基于detectron2的问题还是别的。 感谢解答。 我跑了很多天了，设置多个num_workers后，cpu无法跑满，导致我的速度非常慢，这个怎么办呢。是和你们依托detectron有关吗，还是你别的原因，我看你们在代码中，好像是也有遇到相关问题吗

Could you provide a brief script to show how to evaluate the trained model on YCBV dataset? I just found it is hard to run the evaluation based on your current descriptions. Thx.

sorry to be a bother, when I run generate_pbr_P_fast.py, it shows an error called "'ref' has no attribute 'lm_full'", trace back from line 110, how can I fixed it?

Hi! Thanks for opensourcing the code! I cannot download models by the link: https://drive.google.com/file/d/136ExcMykxsVVSzOiGQVYspq1fx9Hjd6R/view?usp=sharing (error is: "Sorry, the file you have requested does not exist."). Does it work?

您好，我之前在研究gdrnet的相关工作，但一直无法成功运行生成xyz_crop的程序。而我在您开源的这份代码的readme中看到您提及generate p.py对应的是2d-3d matching的groundtruth，这里生成的是否就是gdrnet中需要的xyz_crop呢。如果是的话，我是应该运行generate_pbr_P.py还是generate_pbr_P_fast.py呢？生成的xyz_crop和gdrnet中要求的是否一致呢？

Hi, I have a small question about the results on linemod dataset.

I see that in the paper, the backbone should be ResNet34, however, in the codebase, it seems like ResNet50 (https://github.com/shangbuhuan13/SO-Pose/blob/a3a61d2c97b1084a4754d6c12e45e16d85809729/configs/gdrn_selfocc/lm/gdrn_2rothead_multistep_02CT.py#L53).

I also run the linemod experiments with exact the config file in the repo, getting result ADI.10 ~95.5 with ResNet50. So I would like to confirm with you what is the Backbone used in linemod results?

sorry to be a bother, when I run this python file, it shows an error called "AssertionError: /home/fxj/SO-Pose-main/datasets/BOP_DATASETS/lm/train_pbr/xyz_crop/000000/000000_000001-xyz.pkl", where can I find"000000_000001-xyz.pkl"?